Journal of Information System Research (JOSH)

Comparison of Nazief–Adriani and Porter Stemmer in Determining Javanese Root Words

2026-07-07T18:37:38+07:00

Stemming is an important step in text processing to convert inflected words into their root word. In Javanese, the stemming process is more challenging due to its complex morphological characteristics, including prefixes, infixes, suffixes, and combinations of affixes that are often accompanied by phonological changes. This study aims to compare the performance of the Nazief–Adriani and Porter Stemmer algorithms in Javanese stemming by including infix processing as part of the stemming stage. The dataset used consists of 603 Javanese affixed words covering various types of affixes. The evaluation was conducted using a confusion matrix with accuracy, precision, recall, and F1-score metrics, as well as an analysis of over-stemming and under-stemming errors. The results show that the Nazief–Adriani algorithm performs better with an accuracy of 90.05%, compared to the Porter Stemmer, which achieved 76.12%. This advantage is influenced by the validation of the dictionary at each stage of affix cutting, so that the stemming results are more controlled. The application of infix processing has also been proven to contribute to improving the accuracy of stemming results. This study is expected to be a reference in the development of natural language processing systems for Javanese and encourage further research related to the refinement of morphological rules.

Segmentasi Citra Wayang Kulit Pandawa Berkompleksitas Visual Tinggi Menggunakan Model U-Net Berbasis Convolutional Neural Network

2026-07-07T18:41:16+07:00

Shadow puppetry (wayang kulit) is one of Indonesia's cultural heritages with significant historical and artistic value. The complexity of digital image backgrounds in wayang kulit poses a major challenge in automatic segmentation, particularly due to lighting variations, intricate carving (tatahan) details, and the limitations of conventional methods in handling high visual variability. This study aims to implement a U-Net architecture based on Convolutional Neural Network (CNN) for segmenting images of Pandawa shadow puppet characters encompassing five main characters: Puntadewa, Janaka, Werkudara, Nakula, and Sadewa. The dataset consists of 1,500 independently collected shadow puppet images with ground truth masks divided into 1,093 training, 157 validation, and 250 test data. The U-Net model was trained using the Adam optimizer with an initial learning rate of 1×10⁻⁴, combined Binary Cross-Entropy and Dice Loss function, and 128×128 pixel input size. Early stopping and automatic learning rate adjustment via ReduceLROnPlateau were applied to optimize training and prevent overfitting throughout the learning process. The model achieved Accuracy 95.8%, AUC 98.6%, Dice Coefficient 91.9%, IoU 86.9%, Precision 91.5%, and Recall 95.0% on 250 test data. Previous studies on wayang kulit have been limited to image classification, while U-Net applications have been predominantly found in medical and satellite domains, making this study a novel contribution that addresses an existing research gap and supports the digitalization of Indonesian cultural heritage. The contribution of this study is to provide the first deep learning-based image segmentation model specifically designed to automatically separate Pandawa wayang kulit silhouettes from their backgrounds, demonstrating the effectiveness of U-Net architecture on cultural heritage objects with high visual complexity, and establishing a segmentation performance baseline for the Indonesian visual cultural heritage domain that can serve as a reference for future wayang kulit digitalization system development.

Analisis Sentimen Ulasan Berbahasa Inggris Apex Legends di Steam Menggunakan TF-IDF N-Gram dan Multinomial Naive Bayes

2026-07-07T18:45:25+07:00

The number of users of the online game Apex Legends continues to increase along with the always active community, which also leads to an increase in the number of user reviews. In this condition, conducting manual review analysis becomes ineffective, especially due to the numerous reviews written in informal English, containing negation words, and also showing an imbalanced sentiment class distribution. In this study, the aim is to classify reviews from Apex Legends users on the Steam platform into positive and negative sentiments using the Multinomial Naive Bayes algorithm with TF-IDF weighting based on N-Gram features with a combination of Unigram and Bigram. The dataset was obtained through web scraping from the Steam platform with a total of 9,000 reviews, followed by preprocessing which resulted in 8,981 valid reviews. However, the data still showed class imbalance. The random undersampling process was then applied to obtain 5,512 balanced data points. The test results show that the model can achieve an accuracy of 0.8132 or 81.32%. For the negative class, the model obtained a precision of 0.79, recall of 0.85, and f1-score of 0.82, while the positive class obtained a precision of 0.83, recall of 0.78, and f1-score of 0.81. The trained model is also applied to a Streamlit based dashboard to support the visualization and prediction of new review sentiments. The contributions of this study are the application of combined N-Gram features (unigram and bigram) to Multinomial Naive Bayes for handling negation context and informal language, the use of random undersampling to address class imbalance, and the deployment of the trained model into a Streamlit-based dashboard that enables direct visualization and sentiment prediction of new reviews.

Analisis Kepuasan Pengguna Sistem E-commerce Penjualan Pakaian sebagai Media Pendukung Pembelajaran Menggunakan Metode EUCS

2026-07-07T18:52:41+07:00

The development of digital technology in education has encouraged the use of e-commerce systems as learning support media, making user satisfaction evaluation essential to ensure system effectiveness. This study aims to measure the level of user satisfaction with an e-commerce system utilized in digital learning activities. The research employed a quantitative approach using the End-User Computing Satisfaction (EUCS) method, which consists of five dimensions: content, accuracy, format, ease of use, and timeliness. Data were collected through questionnaires distributed to 61 respondents and analyzed using validity testing, reliability testing, mean analysis, and correlation analysis. The results indicate that all questionnaire items are valid and reliable, while all EUCS dimensions obtained mean scores above 4.000, indicating a high level of user satisfaction. The highest mean values were found in the accuracy and ease of use dimensions (4.268), while correlation analysis revealed that the strongest relationship occurred between the content and format dimensions with a correlation coefficient of 0.726. This study contributes by providing empirical evidence regarding user satisfaction with the use of e-commerce systems as learning media and by identifying the relationships among EUCS dimensions that influence user perceptions. In conclusion, the e-commerce system has successfully met user needs and effectively supported digital learning activities.

Analisis Deskriptif Komparatif Pemanfaatan ChatGPT, Kualitas Pemahaman, dan Efisiensi Tugas Berdasarkan Status Kerja Mahasiswa

2026-07-07T18:58:06+07:00

The use of ChatGPT in academic activities may be perceived differently in relation to students’ understanding and task efficiency. Differences in academic demands between working and non-working students may also lead to different patterns of use. This study aimed to describe ChatGPT Utilization, Quality of Understanding, and Task Efficiency, compare the three constructs according to student employment status, and compare Quality of Understanding and Task Efficiency within the same respondents. A quantitative descriptive-comparative design was employed. Data were collected from 101 Universitas Universal students selected through purposive sampling using a four-point Likert-scale questionnaire. Instrument evaluation resulted in 26 final items across three constructs, with Cronbach’s Alpha values ranging from 0.787 to 0.913. The Mann–Whitney U test indicated no significant differences between working and non-working students in ChatGPT Utilization (p=0.923), Quality of Understanding (p=0.244), or Task Efficiency (p=0.079). The Wilcoxon Signed-Rank Test showed that the mean item score for Task Efficiency was higher than that for Quality of Understanding, at 3.260 and 3.000, respectively (Z=−5.779; p<0.001; r=0.593). Based on respondents’ perceptions, ChatGPT use was more prominent in supporting practical and efficient task completion than in the quality of understanding, while student employment status did not produce meaningful differences across the three constructs. This study contributes empirical evidence by distinguishing utilization, understanding, and efficiency as separate aspects and provides a practical basis for higher education institutions to guide ChatGPT use while maintaining information verification and students’ understanding processes.

Analisis Hubungan Preferensi Genre Musik dan Kesehatan Mental pada Dataset MXMH

2026-07-07T19:05:31+07:00

Mental health is a critical issue that is influenced by various factors, including music-listening habits. In the field of music psychology, music genre preferences are known to be associated with emotional regulation and an individual’s psychological state. This study aims to analyze the relationship between music genre preferences and mental health using a data-driven approach. The dataset used is the Music & Mental Health Survey (MXMH), which consists of 737 respondents with variables including music genre preferences, duration of music listening, and mental health indicators such as anxiety, depression, insomnia, and obsessive-compulsive disorder (OCD). The research stages included data preprocessing, exploratory data analysis (EDA), determining the number of clusters using the Elbow Method and Silhouette Score, clustering using the K-Means algorithm, analyzing the relationship between music genre and mental health, and classifying the clustering results using Random Forest. The results showed that respondents could be grouped into three clusters with distinct mental health characteristics. A Silhouette Score of 0.2246 indicates that the quality of cluster separation is still relatively low, making the segmentation results more exploratory in nature. Correlation analysis revealed a positive relationship between the anxiety and depression variables, as well as differences in music genre preference patterns among groups with different mental health conditions. The feature importance results show that the music genre preference variable contributes to distinguishing the characteristics of each cluster. The contribution of this study is to provide an empirical overview of the relationship between music genre preferences and mental health conditions based on the MXMH dataset through a machine learning-based segmentation and classification approach. The findings of this study suggest that music preferences have the potential to be used as an indicator for understanding patterns of an individual’s psychological state, although further validation and methodological development are needed to achieve a more robust segmentation.

Klasifikasi Mutu Tomat dan Potensi Umur Simpan Berdasarkan Fitur Warna-Tekstur Menggunakan Random Forest

2026-07-07T19:26:10+07:00

Postharvest tomato deterioration remains a major challenge due to manual and subjective quality assessment, which may lead to inconsistent sorting results and inaccurate shelf-life estimation. This study aims to develop a tomato quality classification system and predict potential shelf life based on digital image processing using the Random Forest algorithm. The study employed 936 tomato images and 450 non-tomato images collected independently. The extracted features consisted of Red Green Blue (RGB) and Hue Saturation Value (HSV) color features, as well as Gray Level Co-occurrence Matrix (GLCM) texture features. Tomato quality was classified into three categories, namely Poor, Medium, and Good, using a Random Forest Classifier, while shelf-life prediction was performed using a Random Forest Regressor. The classification model achieved an accuracy of 96.81%, precision of 96.82%, recall of 96.81%, and an F1-score of 96.81%. The regression model produced a Mean Absolute Error (MAE) of 0.0621, a Root Mean Square Error (RMSE) of 0.1152, and an R² value of 0.8752, while cross-validation yielded an average accuracy of 95.83% ± 1.24%, indicating stable model performance. Feature importance analysis revealed that color features contributed the most to both models, with g_mean identified as the most influential feature for tomato quality classification and shelf-life prediction. This study contributes to the development of a tomato quality assessment system capable of simultaneously classifying tomato quality and predicting shelf-life potential based on digital image processing using the Random Forest algorithm. In addition, feature importance analysis is employed to identify the visual characteristics that have the greatest influence on model performance. The results demonstrate that the proposed approach has the potential to support tomato sorting and postharvest management processes in a more objective and efficient manner.

Sistem Informasi Keselamatan dan Kesehatan Kerja Berbasis Web dengan Pemetaan Geografis untuk Deteksi Bahaya dan Pelaporan Insiden di Kampus

2026-07-07T19:30:31+07:00

K3L management in vocational campuses still runs manually. Incident reports are filled out on paper forms, hazard data is stored in separate spreadsheets across units, and there is no single view showing where hazards are located. This study develops a web-based K3L information system for Politeknik Manufaktur Bandung using Laravel 12, MySQL, and Leaflet.js as an interactive GIS map engine, following a Research and Development (R&D) method across eight stages. The system has seven main features: (1) interactive GIS mapping with hazard markers on OpenStreetMap; (2) GPS-based incident reporting via Web Geolocation API with campus polygon boundary validation and accuracy display in meters; (3) hazard reporting with mandatory GPS and building floorplan overlay; (4) voice-to-text using Web Speech API in Bahasa Indonesia across four text fields (chronology, cause, first aid action, and hazard notes); (5) automatic WhatsApp notifications to reporters and task force when reports are submitted or status is updated; (6) GPS location verification by task force on incident reports; and (7) a knowledge center and emergency center accessible without login. Testing used black box testing via Newman (79 requests, 237 assertions, 0 failures), UI smoke testing (22 scenarios, 0 failures), and User Acceptance Testing (14 scenarios from three actors, all accepted). The system transforms the previously fragmented manual K3L workflow into a single centralized digital platform monitored in real-time. This study contributes a practical model through a web-based occupational health and safety information system that integrates interactive GIS mapping, GPS, and voice-to-text in one centralized platform, serving as a replicable reference for K3L digitalization in vocational campuses that can be adapted by other educational institutions.

Visualisasi Area Tanam Perkebunan Berbasis WebGIS Menggunakan Data Foto Udara Resolusi Tinggi

2026-07-07T19:33:50+07:00

Spatial distribution and visualization of planting areas in PTPN IV Regional 4 Kayu Aro Unit were conducted using high-resolution aerial imagery and Geographic Information System (GIS). This study aims to identify the distribution of planting areas and develop a WebGIS as an interactive spatial visualization tool that can support the monitoring and management of plantation land. The methods used include visual interpretation of high-resolution aerial photos, land use digitization, GIS spatial analysis, and WebGIS implementation based on QGIS2Web. Land use classification distinguished planting and non-planting areas, supported by slope analysis to evaluate topographic influence on land utilization. Plantation areas are predominantly occupied by active planting zones, with the highest percentage recorded in Afdeling E (95.881%), followed by Afdeling G (92.735%), Afdeling F (92.689%), and Afdeling D (92.601%). Afdeling A shows the lowest planting proportion at 58.84%, indicating a relatively higher concentration of non-planting areas. Spatial patterns indicate that planting areas are generally distributed on flat to undulating slopes, representing more suitable conditions for cultivation and plantation management. Spatial data and attribute information were integrated into a WebGIS platform to support interactive visualization, spatial monitoring, and information accessibility. The research results show that Afdeling E has the highest planted area percentage at 95.88%, while Afdeling A has the lowest at 58.84%. Black Box testing shows that all WebGIS features work with a 100% success rate.

Peningkatan Keamanan Kunci Vigenère Menggunakan Steganografi Least Significant Bit (LSB) pada Sistem IoT Smart Door QR Code

2026-07-07T19:53:58+07:00

In an increasingly connected digital era, the Internet of Things (IoT) enables seamless real-time data exchange across devices but also introduces critical challenges in data security. Although symmetric cryptography is widely adopted for its computational efficiency, key distribution and protection remain major vulnerabilities. This study aims to enhance IoT network security by integrating the Vigenère Cipher with Least Significant Bit (LSB) steganography. The LSB method is employed to conceal cryptographic keys within digital media, reducing the risk of unauthorized key interception. The proposed system is evaluated across three communication channels: the server, the user web interface, and ESP32/ESP32-CAM devices. Sniffing attack tests confirm that all transmitted data appears only as ciphertext, indicating successful protection of plaintext and secure key exchange. Performance measurements also demonstrate that the combined methods operate efficiently. On the server, encryption and encode require an average of 0.14 ms, while decode and decryption on the user web interface require 0.13 ms. On the ESP32-CAM, encryption and encode average 2.22 ms, with decode and decryption on the server requiring 0.10 ms. For the ESP32, server-side encryption and encode take 0.10 ms, while device-side decode and decryption take 1.46 ms. Overall, the integration of Vigenère Cipher and LSB steganography effectively improves data security in IoT communication without significantly impacting system performance.

Pemanfaataan Penginderaan Jauh dan Sistem Informasi Geografis Berbasis Transformasi Spektral Indeks Vegetasi Untuk Estimasi Produksi Tanaman Teh

2026-07-07T19:59:08+07:00

Tea is a leading plantation commodity in Indonesia, but production estimation through manual field surveys has limitations in terms of cost, time, and accuracy, and is less able to describe spatial variations between blocks representatively. This study aims to estimate tea production in Afdeling B PTPN IV Danau Kembar using PlanetScope imagery with the Transformed Vegetation Index (TVI) approach, and to test the accuracy of the estimation compared to actual production data. The methods used include image pre-processing (radiometric calibration), TVI calculation, field data collection through sample plots, simple linear regression analysis, and production estimation at the block level using zonal aggregation and Jenks Natural Breaks classification. The results show that the TVI value ranges from 0.79–1.13 with a productive land area reaching 240 ha (84.21%) of the total 285 ha. The regression analysis yielded a coefficient of determination (R²) of 0.7933 with the equation y = 1388.8x – 1458.4, while the model validation results showed an R² of 0.7005. The total estimated tea shoot production in Afdeling B reached 119.5 tons, with the highest production in block 24 (8.80 tons) and the lowest in block 46 (0.48 tons). Although the model displayed good accuracy results at the sample scale, the resulting estimate was less precise compared to company data due to differences, namely the temporality of the data. However, the approach using TVI based on PlanetScope imagery has proven to have advantages in presenting spatial information on the distribution of tea plant productivity per block that cannot be obtained from conventional methods, thus supporting more efficient and sustainable spatial data-based tea plantation management. The contribution of this research is to provide a TVI- and PlanetScope-based tea production estimation model applied to the highland tea plantations of West Sumatra, while also generating a spatial productivity distribution map per block as a basis for more practical plantation management decision-making.

Pengembangan Model Deteksi Isu Publik Berbasis Latent Dirichlet Allocation Dengan Pendekatan Tren Waktu dan Analisis Sentimen pada Berita Online Nasional

2026-07-07T20:24:14+07:00

The growth of digital media and online news in Indonesia has generated a massive volume of information that continues to expand daily. This situation makes it difficult to identify public issues quickly and accurately, as manual news monitoring requires significant time, effort, and resources. Furthermore, the multitude of news sources with varying editorial focuses results in fragmented information that is challenging to analyze comprehensively. Consequently, an automated approach is needed to detect and monitor public issues within large datasets of online news. This study aims to develop a public issue detection model for national online news using the Latent Dirichlet Allocation (LDA) method. Research data was obtained via web scraping from CNBC Indonesia, Detik.com, Kompas.com, and Liputan6.com between January and December 2025, yielding 149,335 news headlines; after preprocessing, 146,557 clean data points remained. Topic modeling was performed using LDA, followed by analysis involving temporal trends, spike detection, media comparisons, and sentiment analysis based on the InSet dictionary. The results demonstrate that the LDA model successfully identified 16 key topics representing various public issues. The analysis revealed differences in reporting focus across media outlets, spikes in specific issues during certain periods, and a predominance of negative sentiment across most topics. These findings indicate that the proposed approach is capable of supporting the automated and structured monitoring of public issues.

Kuantifikasi Risiko Introspection pada Tiga Kategori Otorisasi OWASP: Studi Komparatif REST API dan GraphQL

2026-07-07T20:30:46+07:00

The advancement of Application Programming Interfaces (APIs) demands measurable architectural-level security evaluation. This study quantifies the security risks of REST API and GraphQL based on three authorization categories from the OWASP API Security Top 10 2023 (API1, API3, and API5). The exclusive limitation to these three categories was established to focus purely on access control logic flaws rather than infrastructure-level vulnerabilities. The experiment utilizes TixVuln, a parallel-architecture testbed instrument explicitly designed to eliminate external database bias a comparative advantage not present in standard single-architecture vulnerable applications. Authorization evaluation was executed contextually to avoid the high false-negative rates typically produced by automated security scanning tools (SAST/DAST) in business logic testing. Quantification results using the OWASP Risk Rating Methodology reveal a novelty that GraphQL experiences a risk category escalation from Medium to Critical levels in API3 and API5 compared to REST API. This significant leap in the Ease of Discovery metric is absolutely triggered by the operational schema exposure through the introspection feature. Mitigation testing validates that implementing field whitelisting and resolver-level Role-Based Access Control is imperative to suppress inherent risks in single-endpoint architectures. The main contribution of this research is the provision of an isolated empirical evaluation framework that quantitatively proves the flexibility of GraphQL architecture is directly proportional to the increased fatality of authorization risks if the schema discovery feature is not strictly configured.

Pengembangan Aplikasi Web Smart Waste Management Berbasis IoT dan Dashboard Spasial Untuk Optimasi Rute Pengangkutan Sampah

2026-07-14T15:11:37+07:00

Increasing urban waste creates logistical hurdles for the Sleman Waste Services (UPTD). Static collection schedules without actual capacity data lead to operational inefficiencies. This study designs an Internet of Things (IoT)-based Smart Waste Management system integrated with a spatial dashboard to optimize collection efficiency. System development adopted the Waterfall method, encompassing hardware (ESP32 and HC-SR04) and web interfaces. Telemetry transmission utilizes a REST API for real-time visualization. Test results indicated a sensor accuracy of 97.2% with an average network latency of 7.46 seconds. User Acceptance Test (UAT) and black-box tests recorded success rates exceeding 95%. Operationally, the smart routing implementation successfully reduced the daily fleet travel distance from 25 km to 12 km (52% fuel efficiency). The main contribution is twofold. Scientifically, it proposes an integrated architecture combining IoT telemetry and spatial smart routing within a single platform. Practically, the system is proven to reduce fleet travel distance and achieve 52% fuel savings, providing an effective solution to prevent waste overload and realize efficient urban sanitation logistics.

Optimasi Model Retrieval-Augmented Generation Menggunakan Algoritma Indeks HNSW Lokal pada Small Language Model untuk Mitigasi Halusinasi Hadis Bukhari

2026-07-14T16:03:20+07:00

Small Language Models (SLMs) offer high local computational efficiency but possess a systemic vulnerability to information hallucination. This vulnerability becomes a critical risk when the model is applied to sensitive domains demanding absolute accuracy, such as Sharia law and Islamic sacred literature. The inaccurate citation of religious texts can lead to theological misguidance. To address this issue, this study proposes a low-memory, local Retrieval-Augmented Generation (RAG) architectural solution. This system is designed by integrating the open-source Llama-3.2-1B-Instruct model with the PostgreSQL pgvector vector database. Document retrieval optimization is performed based on the HNSW (Hierarchical Navigable Small World) indexing algorithm and a 16-bit precision quantization technique (halfvec) on 7,003 chunks of the Sahih al-Bukhari Hadith corpus. The primary objective of this research is to design a high-precision hallucination mitigation system that operates independently (on-premise), while making a tangible contribution to the development of a low-cost digital theological assistant that preserves privacy and data sovereignty. Mitigation efficacy was automatically evaluated using the Ragas framework against 50 theological test queries, while database efficiency was physically tested on consumer-grade local computer hardware. Experimental results indicate that the proposed architecture is capable of significantly improving the faithfulness metric by 117.4% (from 0.4120 to 0.8960) and answer relevance by 50.0% (from 0.6120 to 0.9180), while simultaneously accelerating inference response time by up to 50.7%. On the database side, halfvec quantization successfully reduced physical table storage space by 36.2% and accelerated index construction time by 16.5% with an absolute accuracy (recall) retention rate of 1.0000. This study proves that a high-precision religious virtual assistant is highly feasible to execute independently without relying on third-party cloud computing services.

Pengembangan Sistem Informasi Early Warning System Banjir Berbasis Internet of Things Menggunakan REST API dan WhatsApp Gateway

2026-07-14T15:37:34+07:00

Flooding is one of the most frequent natural disasters caused by increasing rainfall intensity and overflowing rivers, highlighting the need for a monitoring system capable of providing timely and accurate information. This study aims to develop an Internet of Things (IoT)-based Early Warning System (EWS) for flood monitoring that enables real-time water level observation and automatic early warning notifications. The proposed method involves the design of a hardware system using the A02YYUW ultrasonic sensor and ESP32 microcontroller, the development of a REST API-based backend using PHP Native and MySQL, and a web-based monitoring dashboard integrated with the WhatsApp Gateway through the Fonnte API. Water level data collected by the sensor are processed using a .-based classification method to determine three alert levels: NORMAL, ALERT, and DANGER. The processed data are then transmitted to the server in JSON format via a Wi-Fi network. Subsequently, the backend validates the authentication token, stores the data in the database, displays real-time information on the monitoring dashboard, and automatically sends WhatsApp notifications when the water level reaches the ALERT or DANGER status. The experimental results demonstrate that the proposed system operates successfully in an integrated manner, covering sensor data acquisition, data transmission, database storage, dashboard visualization, and automatic notification delivery. The A02YYUW ultrasonic sensor achieved an average measurement error of 0.89%, indicating high measurement accuracy and stability. Therefore, the developed system can serve as an effective solution for real-time flood monitoring and early warning, providing fast, accurate, and easily accessible information to the community. The main contribution of this study is the development of an integrated flood early warning system that combines the waterproof A02YYUW ultrasonic sensor, ESP32 microcontroller, PHP Native-based REST API, web-based monitoring dashboard, token-based authentication, and WhatsApp Gateway into a unified architecture for secure real-time monitoring and early warning dissemination.

Combining Generative AI and Scheduling Algorithms for Personalized Learning Powered by TELISIK

2026-07-14T15:53:50+07:00

The increasing competitiveness of the Computer Based Written Examination for National Selection Based on Test (UTBK-SNBT) necessitates adaptive and sustainable learning support. This study proposes a web based intelligent learning platform that integrates Artificial Intelligence (AI) to facilitate examination preparation through Automatic Question Generation (AQG), an AI Tutor, automated solution generation, and a scheduler driven question generation mechanism. The platform adopts a client server architecture and employs a Large Language Model (LLM) to generate examination questions tailored to the characteristics of each UTBK-SNBT subtest. The system was evaluated from two perspectives: the learning performance of 30 students across seven UTBK-SNBT subtests and the quality of 1.663 AI generated questions using a Jaccard Similarity based deduplication approach. The results demonstrate that the proposed platform provides continuous and personalised practice beyond conventional static question banks. Acceptable question generation rates reached 96.3% for Quantitative Knowledge, 92.9% for Mathematical Reasoning, and 91.4% for General Knowledge and Comprehension, whereas reading-intensive subtests exhibited higher duplication rates due to the limitations of lexical similarity measurement. Overall, the findings confirm the feasibility of the proposed platform as an intelligent and scalable solution for UTBK-SNBT preparation, while highlighting semantic similarity techniques as a promising direction for improving the quality of text-based question generation.

Design and Development of a Web-Based Student Discipline Evaluation Information System Using a Research and Development Approach

2026-07-14T16:01:22+07:00

Student discipline management is an essential component of educational administration that supports the creation of a conducive learning environment. However, disciplinary management at SMA Negeri 2 Cikarang Selatan was previously conducted manually using notebooks and spreadsheet applications, resulting in data redundancy, inaccurate calculation of violation points, delays in report generation, and difficulties in monitoring students' disciplinary records. This study aims to design and develop a Web-Based Student Discipline Evaluation Information System to improve the efficiency, accuracy, and transparency of disciplinary management. The study employed the Research and Development (R&D) approach, which consisted of needs analysis, system design, system development, implementation, and system evaluation. The proposed application was developed using PHP and MySQL and integrated several functional modules, including user authentication, student data management, violation recording, automatic point calculation, report generation, and user management. Functional evaluation was conducted using the Black-Box Testing method to verify that each module operated according to the specified requirements. The evaluation results indicated that all functional modules performed successfully without significant errors, demonstrating that the developed system satisfied the identified functional requirements. Compared with the previous manual procedure, the proposed system improved administrative efficiency, enhanced data accuracy through automated point calculation, facilitated faster report generation, and increased accessibility to disciplinary information through a centralized web-based platform. The developed system also strengthened communication among administrators, teachers, guidance counselors, and student guardians by providing accurate and timely disciplinary information. Therefore, the proposed Web-Based Student Discipline Evaluation Information System can serve as an effective digital solution for modernizing student discipline management and supporting evidence-based decision-making in educational institutions.

Integrasi Quick Response Dinamis dan Algoritma Geofencing pada Sistem Presensi Terpadu untuk Validasi Kehadiran

2026-07-15T22:17:45+07:00

The integrity of student attendance data in educational settings often faces serious challenges, particularly related to manipulation loopholes such as the use of absenteeism and inconsistencies in attendance locations. In a case study at SMK Kristen Imanuel Laikit, the existing attendance recording method was not fully capable of simultaneously validating location points and identities, potentially triggering administrative fraud. Responding to these problems, this study aims to propose the integration of geofencing and dynamic quick response (QR) algorithms into an integrated attendance system to minimize attendance validation manipulation. Technically, the system generates dynamic QR tokens whose patterns are continuously regenerated every few seconds through a time-based token generation mechanism and a random string combination. This automatic update is strictly designed to prevent reuse of authentication tokens, including the practice of sharing QR code screenshots with other students. Meanwhile, the geofencing mechanism is implemented through the Haversine algorithm calculation to validate the user's device position based on the school's specified radius tolerance limit. This system also integrates an autonomous discipline management mechanism that processes student attendance status based on the timeliness of attendance. Based on software testing results, the system is capable of precisely executing authentication, location validation, QR token synchronization, access rights control, and invalid attendance rejection in an integrated manner. This research contributes to the development of a secure attendance system that is free from location manipulation and QR code duplication. Furthermore, this system is designed to independently manage student discipline to create transparent and accountable academic governance.

Improved Genetic Algorithm with Adaptive Operators and Elitism for Random Forest Feature Selection in Heart Disease Classification

2026-07-15T22:30:19+07:00

Heart disease is one of the leading causes of mortality worldwide, and accurate prediction models are essential to support early diagnosis. However, conventional Random Forest classifiers generally utilize all available features, although not all features contribute equally to classification performance, resulting in unnecessary model complexity. This study proposes an Improved Genetic Algorithm (IGA) that extends the conventional Genetic Algorithm through elitism, adaptive crossover, and adaptive mutation operators to optimize feature selection for Random Forest-based heart disease classification. The proposed method was evaluated using the Cardiovascular Disease Dataset from Kaggle, which consisting of 1,000 records and 14 variables, where 12 predictor features were used for model development. The experimental procedure included data preprocessing, train-test splitting, class imbalance handling using SMOTE on the training set, feature normalization, Random Forest modeling, feature selection using the proposed IGA, and model evaluation. The proposed IGA selected six important features slope, chestpain, restingBP, restingelectro, oldpeak, and gender. The optimized Random Forest model achieved an accuracy of 99.50%, precision of 99.15%, recall of 100.00%, F1-score of 99.57%, and AUC-ROC of 99.90%. These findings indicate that feature selection can simplify the model without compromising classification performance, making the Random Forest + IGA approach a viable alternative for developing more efficient heart disease prediction models.

Perancangan Sistem Absensi Berbasis Internet of Things Menggunakan RFID dan Arduino

2026-07-20T13:10:19+07:00

This study discusses the design of an Internet of Things (IoT)-based attendance system using RFID and NodeMCU ESP8266 to support and improve the efficiency of teacher attendance management at MTs Al-Ikhlas Cicalengka. The primary problem motivating this study is the reliance on a conventional manual attendance system, which is highly vulnerable to recording errors (human error), time inefficiencies during daily data recapitulation, and the risk of physical document loss or damage. Therefore, digital transformation is urgently needed. The proposed system integrates several hardware components: RFID RC522 as the identity card reader, NodeMCU ESP8266 as the main controller and internet (Wi-Fi) connector, and a 16x2 LCD as an interactive information display. The attendance data is directly transmitted and stored in a cloud-based Google Spreadsheet. The system was programmed using Arduino IDE to operate automatically. This study utilized a design and prototyping method by testing 10 registered RFID cards belonging to the teachers. The test results demonstrated that all cards were successfully read, and attendance data were accurately stored in Google Spreadsheet with a 100% success rate. The average system response time was recorded at a very fast 2.2 seconds. In conclusion, the implementation of this IoT-based attendance system is proven to be effective in minimizing data manipulation, accelerating the recording process, and generating attendance recapitulations that are more transparent and easily accessible. The contribution of this research is providing a practical, affordable automation system that can serve as a reference model for other secondary educational institutions in digitally transforming their administrative management.

Rancang Bangun Sistem Informasi Manajemen Barang dengan Fitur Pendukung Keputusan Prioritas Kebutuhan Barang Menggunakan Metode AHP-SAW

2026-07-20T13:19:48+07:00

Inventory management in government agencies often faces obstacles in determining procurement proposal priorities due to manual recording processes and subjective judgment. This study aims to design and build a web-based Inventory Management Information System integrated with a Decision Support System (DSS) using the AHP–SAW method at the Padang City Transportation Office. The system development method used is Waterfall, with the Laravel framework and MySQL database as the technological foundation. The integration of the AHP method is used to determine the criteria importance weights, while the SAW method is used to rank ten operational items alternatives. The results showed that operational urgency became the most dominant parameter with a weight of 0.557. The AHP calculation results were declared consistent with a Consistency Ratio (CR) value of 0.043. The ranking results successfully produced an objective priority sequence of goods requirements based on the combination of criteria weights and alternative values. Functional testing and comparison tests prove that the system produces outputs identical to manual calculations. The implementation of this system has successfully increased warehouse administration efficiency and provided a measurable recommendation instrument for agency leaders in preparing operational goods requirement proposals. The contribution of this study is the integration of an Inventory Management Information System with the AHP–SAW method into a single web-based platform, enabling operational inventory data to be utilized not only for inventory management but also as the basis for determining goods requirement priorities in a more objective and data-driven manner.

Rancang Bangun Sistem Computer Based Test (CBT) Penerimaan Mahasiswa Baru Berbasis Web Menggunakan Metode Research And Development dengan Model Waterfall

2026-07-20T13:34:34+07:00

The implementation of the New Student Admission (PMB) selection process requires an efficient, transparent, and effective system to support accurate decision-making. Previously, the Computer-Based Test (CBT) system used for PMB selection relied on the CodeIgniter framework, which was prone to limitations in data management flexibility, security, and scalability as the number of applicants increased. To address these challenges, this study redesigns a web-based PMB CBT system using the Laravel framework, which adopts the Model-View-Controller (MVC) architecture, and integrates a relational database MySQL to provide structured and maintainable data management. Most previous studies have primarily focused on the digitalization of examination processes, usability improvements, or basic system efficiency. The main innovation and contribution of this research is the implementation of a live score feature that ensures transparent and immediate presentation of examination results for participants. This study employs the Research and Development (R&D) method to design, develop, and evaluate the proposed system. The developed system includes several key features, such as participant management, question bank management, online examinations, automatic score calculation, and a live score feature for real-time result monitoring. Based on Black Box Testing, the system achieved a functional validity rate of 100%, indicating that all testing scenarios were successfully executed without errors. The proposed system contributes by enhancing the transparency of score calculation and utilizing a scalable technology architecture to assist the admission committee in managing large-scale applicant data efficiently.

Sentiment Analysis Kepercayaan Publik Terhadap Pertamina Pada Media Berita di YouTube

2026-07-20T14:09:46+07:00

Public trust in PT Pertamina (Persero) is reflected in public opinion expressed on social media, particularly through the comment sections of news videos on YouTube. This study aims to analyze public sentiment toward Pertamina based on comments posted on YouTube news videos in order to identify the distribution of positive, negative, and neutral opinions. The data were collected using the YouTube Data API through a web crawling process and subsequently processed using several text preprocessing techniques, including cleaning, case folding, word normalization, tokenization, and stopword removal. Sentiment labeling was performed automatically using the VADER Sentiment method, which classified the comments into three categories: positive, negative, and neutral. Feature extraction was then conducted using the Term Frequency–Inverse Document Frequency (TF-IDF) method. The classification process compared the performance of three Naïve Bayes variants, namely Gaussian Naïve Bayes, Multinomial Naïve Bayes, and Bernoulli Naïve Bayes. Of the 56,868 comments that were successfully crawled, 22,527 comments were successfully collected. The sentiment labeling results revealed that neutral sentiment dominated the dataset, accounting for 95.89% of all comments, followed by positive sentiment at 2.86% and negative sentiment at 1.25%. The experimental results demonstrated that Multinomial Naïve Bayes achieved the best performance with an accuracy of approximately 95%, outperforming Bernoulli Naïve Bayes (approximately 92%) and Gaussian Naïve Bayes (approximately 79%). This superior performance is attributed to the compatibility of the Multinomial Naïve Bayes algorithm with TF-IDF feature representation, which is based on word frequency. The findings of this study are expected to provide valuable insights for Pertamina and policymakers in formulating more responsive and effective public communication strategies.

Sentiment Analysis of YouTube Comments on Indonesia-U.S. Trade Agreement: A Comparison of Machine Learning and Deep Learning

2026-07-30T21:55:34+07:00

The Indonesia–United States Trade Agreement has sparked much public debate on YouTube, but due to the large volume of data and the informal nature of the text, manual analysis is ineffective. Through a comparative study of machine learning (Naïve Bayes and Support Vector Machine) and deep learning (Long Short-Term Memory and IndoBert), this research aims to identify the polarity of public opinion and evaluate the best computational model. This study contributes a comprehensive empirical comparison of the four algorithms within an identical experimental pipeline on an issue that has not been previously explored, thereby offering methodological insights for future sentiment analysis research on similarly informal, domain-specific social media text. The dataset consists of 2,284 comments labeled using the InSet Lexicon. The analysis results show that the neutral class dominates the sentiment distribution (52,0%), followed by the positive class (20,2%) and the negative class (27,8%). Performance evaluation reveals that IndoBERT significantly outperforms other models with an accuracy of 87.71% and a Macro F1-Score of 0.87. Support Vector Machine ranked second (79.00%), followed by Naïve Bayes (61.71%). In contrast, Long Short-Term Memory failed completely (accuracy of 52.00%) due to the “majority class collapse” phenomenon in small-scale datasets. This study concludes that the Transformer architecture (IndoBert) is the most robust approach for classifying sentiment in informal Indonesian-language text containing specific geopolitical and economic terms.

Prediksi Promosi Pegawai dengan Stacking Ensemble dengan SMOTE-ENN dan SHAP

2026-07-20T14:25:18+07:00

The paradigm of human resource management in the digital era demands an objective and data-driven employee promotion process. However, the extreme class imbalance (ratio 10.74:1) has the potential to introduce bias against minority groups that deserve promotion. This study proposes a stacking ensemble classification framework consisting of Random Forest, XGBoost, and LightGBM as base learners and Logistic Regression as a meta-learner, with the integration of SMOTE-ENN and two-level SHAP interpretability. This study shows that the application of SMOTE-ENN before cross-validation can result in a biassed performance estimate of up to +110% on the F1-Score; thus, the use of imblearn.Pipeline is proposed, which restricts resampling only to the training fold. Based on the evaluation using 10-fold stratified cross-validation free from data leakage, the stacking ensemble model achieved an accuracy of 0.9022, precision of 0.4342, recall of 0.4889, F1-score of 0.4598, and AUC-ROC of 0.8053. Although it did not achieve the highest F1 score, this model attained the best recall value among competitive models, making it relevant for contexts sensitive to false negative errors. SHAP analysis identifies avg_training_score, age, and performance_index as the main determinants of promotion decisions. The proposed framework provides a methodological contribution to model evaluation on imbalanced data while offering a more transparent and accountable decision support system to support the implementation of meritocracy in both government and corporate organisations.

Implementasi Naïve Bayes Classifier pada Sistem Informasi untuk Klasifikasi Ketepatan Kelulusan Santri

2026-07-20T14:33:21+07:00

Pondok Pesantren Mahasiswa (PPM) is an educational model that integrates higher education with the Islamic boarding school system to produce graduates with strong academic competence and religious character. However, the dual demands of academic and boarding school activities increase the risk of delayed graduation among students, highlighting the need for early intervention to prevent study failure. This study aims to develop a web-based system that classifies students' on-time graduation using the Naïve Bayes algorithm. The dataset was obtained from PPM archives and consisted of 211 student records with attributes including gender, region of origin, study duration, Al-Qur'an achievement, and Al-Hadith achievement. The proposed system classified 172 students as likely to graduate on time and 39 students as likely to graduate late. Performance evaluation using a confusion matrix achieved an accuracy of 86.79%, a precision of 93.41%, and a recall of 91.40%, indicating good classification performance. Furthermore, User Acceptance Testing (UAT) showed high user satisfaction, with acceptance rates of 95.66% from students and 84% from boarding school administrators, both categorized as excellent. The main contribution of this study is the integration of academic and Islamic boarding school attributes, particularly Al-Qur'an and Al-Hadith achievements, into a Naïve Bayes classification model and its implementation in a web-based system to support monitoring, evaluation, and decision-making related to students' on-time graduation.

Optimasi Rute Distribusi Spare Part Motor Menggunakan Ant Colony Optimization untuk Efisiensi Jarak Tempuh

2026-07-20T14:50:45+07:00

Logistics efficiency in goods distribution has become a major challenge for distribution companies in reducing operational costs. This study aims to optimize the distribution routes of motorcycle spare parts at PD. Tri Jaya Motor using the Ant Colony Optimization (ACO) algorithm. The main issue faced by the company is the use of unsystematic manual route determination, which results in inefficient travel distances. The methodology employed in this research involves modeling the Traveling Salesman Problem (TSP) across six distribution points using parameters of alpha = 1, beta = 1, and rho = 0.1. The simulation results demonstrate that the ACO algorithm successfully identified the optimal route with a total distance of 206.6 km, resulting in significant savings compared to the initial route. This study contributes by providing a metaheuristic-based decision-making strategy for medium-scale distribution systems.

Pengembangan Website Responsive Sebagai Portal Informasi Kegiatan dan Berita Himpunan Mahasiswa Menggunakan Metode Research And Development dengan Model Waterfall

2026-07-20T15:25:10+07:00

The dissemination of activity information and news within Student Associations is often carried out through various separate platforms, resulting in information fragmentation and reducing the effectiveness of information delivery to students. This condition makes it difficult for students to obtain information due to the increased risk of missing important organizational activities and news. Therefore, a centralized information platform is needed to provide integrated information that can be easily accessed through various devices. This research aims to design and develop a responsive website as a centralized portal for Student Association activities and news using the Next.js framework. This study applies the Research and Development (R&D) method with the Waterfall development model, which consists of several stages, namely requirements analysis, system design, implementation, and testing to ensure that the website operates in accordance with user requirements. The system was developed using Next.js as the main framework and Tailwind CSS to support the implementation of Responsive Web Design, enabling the website to be accessed optimally on desktop, tablet, and smartphone devices. This research resulted in a system that successfully integrates news, activities, organizational profiles, and documentation into a single centralized digital platform. System testing was conducted using the Black Box Testing method by evaluating all major functions based on predetermined input and output scenarios without examining the internal structure of the program code. The test results showed that all functions operated in accordance with the specified functional requirements, achieving a 100% success rate across all testing scenarios. The developed website enables students to access centralized information on Student Association activities and news more conveniently through a wide range of devices.

Analisis Kontribusi Sensor IoT pada Deteksi Kebakaran Lahan Gambut Menggunakan Random Forest dan SHAP

2026-07-20T15:33:28+07:00

Peatland fires are disasters that impact the environment, health, and socio economic activities. Internet of Things (IoT) based early detection enables real-time monitoring of environmental conditions through various sensors. However, the specific contribution of each sensor to the detection process remains unclear. This study aims to analyze the contribution of multiple sensors within an IoT-based peatland fire detection system using the Random Forest (RF) algorithm. The dataset comprises 2,000 primary data points obtained from AMG8833, MAX6677, MQ-2, DHT22, and Water Float sensors. The model was trained on the primary data and tested against data representing transitional (overlapping) conditions between normal states and fire events. Model performance was evaluated using accuracy, precision, recall, F1-score, and a confusion matrix, while sensor contributions were analyzed via Feature Importance and validated using SHapley Additive exPlanations (SHAP). The results indicate that the RF model achieved an accuracy of 87.50%, precision of 100.00%, recall of 75.00%, and an F1-score of 85.71%. Feature Importance and SHAP analyses revealed that the DHT22 sensor (measuring humidity and temperature) made the most significant contribution, followed by the MAX6677, MQ-2, AMG8833, and Water Float sensors. These findings demonstrate that temperature and humidity serve as key indicators for peatland fire detection and provide a foundation for developing IoT-based detection systems.

Klasifikasi Penyakit Daun Bawang Merah Menggunakan MobileNetV2 Convolutional Neural Network (CNN)

2026-07-21T08:49:06+07:00

Diseases affecting shallot plants are a primary cause of reduced crop quality and yield. Manual disease identification relies on visual observation, making it prone to error and time-consuming. This study aims to develop a classification model for shallot leaf diseases by combining Gray Level Co-occurrence Matrix (GLCM) feature extraction with MobileNetV2, classified using a Convolutional Neural Network (CNN). The dataset comprises 1,188 shallot leaf images categorized into five classes: downy mildew, healthy, leaf blight, *moler* (basal rot), and purple blotch. The research process involved dataset collection; pre-processing (image resizing to 224×224 pixels, grayscale conversion, normalization, and data augmentation); texture feature extraction using GLCM; and deep feature extraction using MobileNetV2. These features were then combined and used as input for the CNN classification model. Model evaluation was conducted using a confusion matrix, assessing accuracy, precision, recall, and F1-score. The results demonstrate that the proposed model achieved 93% accuracy, with balanced precision, recall, and F1-score values across most classes. The contribution of this research is the integration of gray level co-occurrence matrix (GLCM) texture feature extraction with mobilenetv2 visual features within a convolutional neural network (CNN) model to improve the classification performance of shallot leaf diseases.

Deteksi Kantuk Pengemudi Berbasis Eye Aspect Ratio dan Head Pose Estimation dengan Integrasi IoT

2026-07-21T09:05:41+07:00

This study proposes a real-time driver drowsiness detection system integrating computer vision with Internet of Things (IoT) technology on an affordable embedded platform. The system uses a Camera Module 3 Wide NoIR connected to a Raspberry Pi 3 Model B+ to capture driver facial images. Two visual indicators are computed in parallel from 68 facial landmarks extracted using dlib: Eye Aspect Ratio (EAR) for detecting prolonged eye closure, and Head Pose Estimation using solvePnP for detecting head nodding. An OR-logic decision mechanism triggers an audio alarm when EAR falls below 0.25 for five consecutive frames or Pitch angle exceeds 15 degrees for ten consecutive frames. Events are classified as KANTUK_MATA, KANTUK_KEPALA, or KEDUANYA and sent to firebase Realtime Database for remote monitoring. Black Box testing with 11 scenarios confirms all core functions operate correctly. Average response times of 2.00 seconds via EAR and 2.70 seconds via Head Pose are within acceptable ranges for early drowsiness warning. The multi-indicator approach demonstrates that head nodding is detected earlier than eye closure in gradual drowsiness scenarios, providing earlier warning than single-indicator systems. Under high CPU load conditions (>80%), the frame rate drops to 5 fps, resulting in a total system latency of 2.8–4.2 seconds; this condition is still adequate for early warning but not for sub-second responses. Across the 11 Black Box testing scenarios, the system achieved 100% precision (no false alarms during normal blinking) and 100% recall (all drowsiness conditions detected), although the testing was conducted under controlled laboratory conditions, necessitating further generalization to real-world environments.

Implementasi Sistem Informasi Pengelolaan Surat Berbasis Web dengan Metode Rational Unified Process pada Politeknik

2026-07-21T09:33:51+07:00

Management of incoming and outgoing mail at Indotec Kendari Polytechnic is still done manually through recording in an agenda book, so the archive search process is less efficient, risks causing loss or damage to documents, and slows down the preparation of administrative reports. This study aims to implement a web-based mail management information system to improve the efficiency of the process of recording, storing, searching, and reporting letters. The system development method used is the Rational Unified Process (RUP) which consists of the stages of inception, elaboration, construction, and transition. The system was developed web-based using the PHP programming language and MySQL database, then tested using the Black Box Testing method to evaluate the suitability of the system functions with user needs. The test results show that all main functions of the system, including user authentication, incoming mail management, outgoing mail management, archive search, document upload, and report preparation, run according to functional requirements with a test success rate of 100%. The system implementation is able to improve the effectiveness of mail archive management, speed up the document search process, and simplify the preparation of administrative reports. The contribution of this research is to produce a web-based mail management information system developed using the Rational Unified Process approach, thus serving as an alternative digital solution to support the transformation of mail administration in higher education environments.

Sistem Pendukung Keputusan Penentuan Lokasi Early Warning System Tanah Longsor Menggunakan TOPSIS

2026-07-21T09:55:36+07:00

As one of the areas in the Special Region of Yogyakarta with a relatively high level of landslide vulnerability, Gunungkidul Regency requires targeted mitigation measures, including implementing an Early Warning System (EWS) to detect potential disasters early. The obstacle faced by the Gunungkidul Regency Regional Disaster Management Agency (BPBD) is the absence of a system that can support the process of determining priority locations for EWS installation objectively by considering various existing risk parameters. This research aims to develop a Decision Support System (DSS) by applying the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) method to produce priority recommendations for landslide EWS installation locations. The study utilized historical landslide occurrence data from BPBD Gunungkidul Regency covering the period of 2022–2025 and the Gunungkidul Regency Disaster Risk Assessment Document 2026–2030, with 116 villages (kalurahan) as the decision alternatives. The evaluation was conducted using four criteria: hazard, vulnerability, capacity, and landslide occurrence frequency. The results indicate that Sawahan Village achieved the highest preference value of 1.0000, followed by Mertelu Village with 0.9336 and Ngalang Village with 0.7481, making them the highest-priority locations for landslide EWS installation. The developed system successfully automates the TOPSIS calculation process, provides ranking results in a fast and transparent manner, and fulfills all functional requirements based on Black Box Testing. The proposed system is expected to support BPBD Gunungkidul Regency in determining priority locations for landslide Early Warning System (EWS) installation through a more measurable, data-driven, and objective decision-making process.

Automating Clean Architecture Conformance via AST-Based Code Smell Detection and Auto-Refactoring in Laravel Projects

2026-07-21T10:16:38+07:00

While MVC frameworks like Laravel accelerate development, their inherent architectural flexibility frequently induces structural code smells such as fat controllers and direct database access that inevitably lead to architectural erosion and technical debt. This research proposes a Visual Studio Code (VS Code) extension designed to automate Clean Architecture conformance through real-time Abstract Syntax Tree (AST) parsing. Utilizing the Tree-sitter parsing system, the extension continuously evaluates the codebase to detect architectural violations. Beyond passive diagnostics, the tool facilitates interactive auto-refactoring. To ensure business logic integrity during migration, the refactoring engine employs deterministic AST node transformations that safely isolate data access operations into dedicated service layers without altering operational behavior. Evaluation through Black-Box Testing confirmed high detection accuracy across predefined architectural anti-patterns. Furthermore, User Acceptance Testing (UAT) conducted with industry experts yielded an average acceptance score of 4.18 out of 5.00. Ultimately, this AST-based approach provides a pragmatic solution to actively prevent architectural decay, mitigate technical debt, and significantly enhance the long-term maintainability of Laravel-based projects. The main contribution of this research is the transition of architectural code smell mitigation from a passive diagnostic reporting mechanism into an active, real-time, AST-driven auto-refactoring tool directly integrated within the developer's workspace.

Optimasi Pengenalan Plat Nomor Kendaraan Berbasis Mobile Menggunakan Google ML Kit: Implementasi dan Analisis Akurasi

2026-07-21T10:22:36+07:00

Manual vehicle identification in parking systems and access control is still slow, error-prone, and inefficient, while the number of vehicles in Indonesia continues to increase. This study aims to design and implement a mobile-based vehicle license plate recognition system using Google ML Kit Text Recognition on the Android platform. Google ML Kit is used as an OCR SDK with a pre-trained model; therefore, this study does not retrain a CNN model and does not claim to develop a deep learning architecture from scratch. The contribution of this study is to provide an Android-based license plate OCR implementation workflow, integrate recognition results with a Supabase database, and evaluate character-level accuracy on license plate images. The system workflow consists of license plate image acquisition, lightweight client-side preprocessing, on-device text recognition, OCR result normalization, database matching, and character-level accuracy evaluation. The test data consist of 25 license plate images collected from a public Kaggle dataset. The results show that 20 out of 25 plates were read perfectly, while five samples contained partial recognition errors. Of 201 tested characters, 194 were correctly recognized, resulting in an average character recognition accuracy of 96.5%. Recognition errors were mainly affected by similar character shapes, image quality, light reflection, capture angle, and blur. These results indicate that Google ML Kit can be applied as a practical mobile OCR solution, although the validation remains limited by the small number and limited variation of test samples.

Implementasi Metode Case-Based Reasoning dengan Pembobotan Gejala untuk Diagnosa Penyakit Kulit pada Kucing

2026-07-21T10:54:02+07:00

Skin diseases are common health issues in cats that can diminish their quality of life if not addressed early. Limited access to veterinarians makes it difficult for many cat owners to initially identify the diseases affecting their pets. Early diagnosis is challenging because many skin diseases present overlapping symptoms, making it hard for owners to distinguish between conditions without professional veterinary assistance. Therefore, this study aims to develop an expert system for the early diagnosis of these conditions using the Case-Based Reasoning (CBR) method. The system's knowledge base was constructed from a collection of 14 disease types and 36 symptoms, derived from a literature review and validated by experts. To address symptom overlap, the CBR method was implemented by calculating the similarity level between a new case and the existing case base using symptom weighting; this allows the system to differentiate between potential diseases based on similarity scores, even when multiple diseases share the same symptoms. The system then selects the case with the highest similarity score as the preliminary diagnosis. The system was developed using the Python programming language and an SQLite database. Key features include symptom selection, a diagnostic process, disease information, and the ability to generate reports in PDF format. Functional accuracy testing conducted on 14 test cases showed that 13 cases were correctly diagnosed against reference data, resulting in a functional accuracy rate of 92.85% and an average similarity score of 77.33%. The study contributes by optimizing symptom weighting within the CBR algorithm to resolve the issue of overlapping disease symptoms and by providing an integrated system that delivers disease information, care tips, similarity scores, and diagnostic reports. The results demonstrate that the CBR method is a viable approach for the early diagnosis of feline skin diseases and can serve as a helpful resource for cat owners to gather information before seeking further veterinary consultation.

Optimalisasi Hyperparameter Random Forest Menggunakan Random Search untuk Klasifikasi Risiko Stunting Pada Balita

2026-07-24T14:31:56+07:00

Stunting is a condition of impaired child growth resulting from chronic nutritional deficiency over an extended period. This condition not only hinders physical growth but also impedes learning abilities and increases the risk of various future diseases. The high prevalence of stunting in Bima City highlights the need to utilize Machine Learning methods to analyze stunting risk factors more accurately. This study aims to optimize the performance of the Random Forest algorithm using the Random Search method to classify stunting risk among children under five in Sambinae Urban Village, Bima City. The dataset comprises records for 1,162 children under five, featuring 20 attributes obtained from the Mpunda Community Health Center (Puskesmas) in Bima City. The research stages include data collection, preprocessing, data splitting, construction of a baseline Random Forest model, hyperparameter optimization using Random Search, evaluation via a Confusion Matrix (based on Accuracy, Precision, Recall, and F1-Score), and feature importance analysis. Prior to optimization, the baseline Random Forest model yielded an accuracy of 70,39%, precision of 56,67%, recall of 62,96%, and an F1-score of 59,65%. Following optimization with Random Search, model performance improved to an accuracy of 71,67%, precision of 58,43%, recall of 64,20%, and an F1-score of 61,18%. The results demonstrate that hyperparameter optimization using Random Search effectively enhances the Random Forest model's performance in classifying stunting risk. The study contributes a stunting risk classification model based on the Random Forest algorithm, optimized via Random Search to achieve a more effective hyperparameter combination than the default settings. Furthermore, the study provides a comparative performance analysis before and after optimization, along with insights into the variables that most significantly influence stunting risk classification. These findings are expected to assist healthcare professionals and local government authorities in identifying stunting risks more rapidly and accurately, thereby serving as a foundation for formulating more targeted prevention strategies.

Smart Story Archive for WhatsApp Status: A Design Thinking-Based Model for Private Status Retrieval

2026-07-24T14:33:08+07:00

WhatsApp Status supports quick sharing of daily moments, yet its 24-hour visibility creates a gap between temporary public sharing and private long-term retrieval. Existing story features mainly emphasize posting and viewing, while private retrieval by date, person, location, keyword, media type, favorite mark, and download access remains limited in WhatsApp Status. This study proposes Smart Story Archive as a user-centered design concept for private Status retrieval. The method used Design Thinking with survey-based needs analysis involving 100 experienced WhatsApp users, selected because they actively used WhatsApp Status and could evaluate archive-related pain points. This study contributes a theoretical and empirical framework for ephemeral data retrieval in private messaging environments by introducing a conceptual model that reconciles user desires for both ephemerality and long-term memory accessibility. The data were analyzed using descriptive statistics and translated into feature requirements and a high-fidelity prototype in Google Stitch. The key findings show three dominant needs: users want to recover meaningful Status memories, retrieve content through incomplete memory cues, and maintain privacy control over archived metadata. The proposed design addresses these needs through multi-cue search, optional private archiving, metadata control, and deletion access. This study contributes a WhatsApp-specific archive model that balances memory retrieval, privacy, and interface simplicity for future usability testing.

Segmentasi Perilaku Pemustaka Menggunakan DBSCAN untuk Optimalisasi Layanan Perpustakaan Digital

2026-07-24T14:42:39+07:00

The rapid growth of digital libraries has generated increasingly large borrowing transaction data, creating the need for analytical techniques to understand user behavior patterns and support data-driven library management. This study aims to cluster library users based on their borrowing behavior using the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm to identify user segments according to their borrowing activity. The study employed the Book-Crossing Dataset, consisting of 278,858 rating transactions, 271,379 user records, and 271,360 book records. The research methodology included Exploratory Data Analysis (EDA), data preprocessing, feature engineering, feature standardization using StandardScaler, ε parameter selection through the K-Distance Graph, DBSCAN clustering, cluster evaluation using the Silhouette Score, and visualization using Principal Component Analysis (PCA). The experimental results indicate that ε = 0.5 and MinPts = 5 produced three clusters, with 244 users identified as noise. The obtained Silhouette Score of 0.5996 demonstrates a reasonably good clustering quality. Furthermore, the resulting clusters successfully represent users with low, moderate, and very high borrowing activities, providing valuable insights for developing personalized library services, improving book recommendation systems, supporting collection development, and facilitating data-driven decision-making to enhance the overall quality of digital library services.

Prediksi Indeks Pembangunan Manusia Menggunakan Support Vector Regression dengan Optimasi Particle Swarm Optimization

2026-07-24T14:48:15+07:00

The Human Development Index (HDI) is a key indicator for measuring regional development performance and serves as an essential reference for evidence-based policy formulation. Accurate HDI prediction is crucial to support effective development planning and decision-making. This study aims to develop an HDI prediction model using Support Vector Regression (SVR) optimized with Particle Swarm Optimization (PSO) to improve prediction accuracy. The dataset was obtained from Statistics Indonesia (BPS), covering 38 provinces during the 2015–2025 period with a total of 421 observations. The research process consisted of data preprocessing, Min-Max Scaling normalization, an 80:20 train-test split, SVR model development, parameter optimization using PSO, and performance evaluation based on Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and the coefficient of determination (R²). The results show that the baseline SVR model achieved an MAE of 0.069187, RMSE of 0.093548, and R² of 0.492454. After PSO optimization, the model performance improved, achieving an MAE of 0.060864, RMSE of 0.084224, and R² of 0.588583. These findings demonstrate that PSO effectively enhances the predictive performance of SVR by identifying optimal parameter combinations. The main contribution of this study is the development and validation of an optimized SVR-PSO framework for HDI prediction using multi-provincial socioeconomic data in Indonesia, providing a more accurate machine learning-based approach to support data-driven human development planning and policy formulation.

Clustering Pola Penggunaan Energi pada Smart Home Menggunakan DBSCAN Berbasis Data Time Series Sensor

2026-07-24T14:56:02+07:00

The increasing adoption of Internet of Things (IoT) devices in smart homes has generated continuous and complex energy consumption data, requiring effective clustering techniques to identify household energy usage patterns. This study aims to cluster household energy consumption patterns using the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm based on sensor time-series data. The study utilized the Smart Home Energy Consumption Dataset, consisting of approximately 90,000 observations with six main variables: Energy Consumption, Peak Hours Usage, Household Size, Average Temperature, Has AC, and Weekday. The research workflow included feature selection, data cleaning, data normalization using StandardScaler, parameter determination through the K-Distance Graph, DBSCAN clustering, and clustering evaluation using the Silhouette Score. Experimental results indicated that the optimal parameters were ε = 0.38 and MinPts = 5, producing 89 clusters, 541 noise observations (0.60%), and a Silhouette Score of 0.0690. Cluster characteristic analysis revealed that energy consumption, peak-hour energy usage, air conditioner ownership, household size, and ambient temperature were the primary factors distinguishing household energy usage patterns. The findings demonstrate that DBSCAN effectively identifies household energy consumption patterns while detecting outliers without requiring the number of clusters to be predefined, making it a promising approach for supporting intelligent energy management systems in smart home environments.

Analisis Performa Waktu Eksekusi dan Penggunaan Memori Algoritma Linear Search dan Binary Search pada Sistem Pencarian Buku Perpustakaan Digital

2026-07-24T15:04:37+07:00

The growth of digital libraries has increased the need for book search systems capable of providing fast and efficient retrieval as the amount of stored data continues to grow. Although Linear Search and Binary Search algorithms have been widely applied in search processes, their implementation and performance evaluation in web-based digital library systems still require analysis under practical implementation conditions. This study aims to analyze and compare the performance of both algorithms based on execution time and memory usage in a digital library book search system. An experimental method was employed by implementing Linear Search and Binary Search using PHP and MySQL. Performance testing was conducted on datasets containing 10, 50, 100, 500, and 1000 book records under the same testing environment. The results indicate that Binary Search achieved lower execution times for most dataset sizes; however, its execution time was slightly higher than Linear Search when tested with 1000 records. The average memory usage of Linear Search was 0.000048 MB, while Binary Search averaged 0.000043 MB, although Binary Search consumed more memory on the 500- and 1000-record datasets. These findings demonstrate that algorithm performance is influenced not only by theoretical complexity but also by system implementation, data characteristics, and the testing environment. Therefore, selecting a search algorithm should consider the implementation context and system requirements to achieve optimal performance.

Prediksi Penyakit Jantung Berbasis Random Forest dengan GridSearchCV dan SHAP Menggunakan Dataset Publik UCI Cleveland

2026-07-27T09:12:23+07:00

Cardiovascular disease remains the leading cause of death globally, with 19.2 million fatalities recorded in 2023, making the development of automated data-driven prediction systems an urgent necessity. This study develops a heart disease prediction model based on Random Forest optimized using Grid Search Cross-Validation (GridSearchCV) and supplemented with SHapley Additive exPlanations (SHAP) analysis for clinical interpretability on the UCI Cleveland Heart Disease dataset (297 samples, 13 clinical features). Three methodological contributions are implemented: (1) a reproducible preprocessing pipeline with post-split StandardScaler to prevent data leakage, (2) deterministic and exhaustive hyperparameter search using GridSearchCV with 216 combinations and Stratified 10-Fold Cross Validation, and (3) SHAP analysis at the global level based on training data and at the local level based on test data to produce clinically interpretable predictions. The optimal hyperparameter configuration obtained is n_estimators = 100, max_depth = None, min_samples_leaf = 4, min_samples_split = 2, and max_features = 'sqrt'. The Tuned RF model achieves an AUC-ROC of 0.9453 and CV AUC of 0.9010, outperforming the RF Baseline (AUC-ROC 0.9414; CV AUC 0.8804) on key discrimination metrics with greater stability. SHAP analysis identifies cp (mean |SHAP| = 0.1031), thal (0.0976), and ca (0.0805) as the three most influential clinical features, consistent with established cardiological diagnostic indicators. The integration of GridSearchCV and SHAP produces a model that is not only accurate but also transparent in supporting medical decision-making.

Implementasi MobileNetV2 Pada Aplikasi Forensik Android Untuk Deteksi Citra AI-generated dengan Ketahanan Terhadap Transformasi Citra

2026-07-27T09:20:28+07:00

The development of Generative Artificial Intelligence has produced realistic synthetic images that are difficult to distinguish from authentic images through visual inspection. This study aims to implement an Android-based mobile digital forensics application for detecting AI-generate d images using MobileNetV2 converted to TensorFlow Lite for on-device inference. A quantitative-experimental approach used 2,000 images, consisting of 1,000 authentic and 1,000 AI-generate d images. The authentic class comprised 500 smartphone photographs and 500 GenImage samples, while the AI-generate d class included Stable Diffusion v1.4, Stable Diffusion v1.5, Wukong, and Midjourney images. The dataset was divided into 1,000 training, 500 validation, and 500 testing images. Training data were used for model development, validation data for model selection and threshold determination, and testing data only for final evaluation. Robustness testing used copies of 500 testing images without retraining. Under normal conditions with a threshold of 0.680, the model achieved 85.40% accuracy, 85.53% precision, 85.40% recall, and an 85.39% F1-score. JPEG q=65 compression produced 86.00% accuracy. The largest degradation occurred with 112 × 112 resizing combined with JPEG q=65, resulting in 62.40% accuracy, a decrease of 23.00 percentage points, and a 56.66% F1-score. The application performed local inference, displayed prediction labels and confidence scores, and stored detection history. This study contributes an on-device detection application and robustness evaluation using a consistent test subset, positioning the system as an initial detection aid rather than a final forensic verification tool.

Efektivitas dan Keandalan Protokol MQTT pada Internet of Things: Tinjauan Literatur Sistematis Berbasis PRISMA 2020

2026-07-27T09:33:28+07:00

This study aims to present a systematic literature review (SLR) of the performance, security, and Quality of Service (QoS) of the Message Queuing Telemetry Transport (MQTT) protocol in Internet of Things (IoT) environments based on articles published between 2020 and 2026. The SLR follows the PRISMA 2020 guidelines and the Kitchenham and Charters guidelines for software engineering studies. A total of 30 core articles, comprising 7 reputable international journals and 23 nationally accredited SINTA journals, were purposively selected based on thematic relevance and methodological quality. The search was conducted using IEEE Xplore, ScienceDirect, SpringerLink, MDPI, Garuda, and SINTA with the keywords “MQTT,” “IoT,” “performance,” “security,” and “QoS.” The selection process followed the PRISMA flow, consisting of identification (n=487), screening (n=86), eligibility assessment (n=43), and included studies (n=30). The data were extracted and thematically synthesized to address four research questions (RQ1–RQ4), with Scopus-indexed international studies included as contextual comparators. The findings indicate that MQTT outperforms HTTP in terms of bandwidth efficiency and low latency, particularly on resource-constrained devices, with reported latency ranging from 45 to 210 ms depending on the QoS level and testing scenario. The dominant performance parameters include latency, throughput, and packet loss. Security approaches encompass TLS/RSA encryption, domestic cryptography, and Random Forest-based attack detection, achieving an accuracy of 97.2%, although TLS encryption may reduce throughput by up to 30%. QoS 1 represents a balanced option for non-critical applications, whereas QoS 2 is appropriate for vital data. The main research gap is the absence of integrated performance, security, and QoS testing within a single experimental framework. Therefore, future research should focus on lightweight security, context-aware adaptive QoS, and real-world-scale validation through integrative studies that examine the interaction among these three aspects.

Sistem Rekomendasi Produk Sparepart Motor Menggunakan Metode Knowledge Based Recommendation

2026-07-27T09:43:41+07:00

The rapid growth of e-commerce has led to an increasing number of motorcycle spare part products available across various online marketplaces. This condition often makes it difficult for users to select products that match their vehicle type, needs, and budget. This study aims to develop a motorcycle spare parts recommendation system using the Knowledge-Based Recommendation method to assist users in obtaining suitable product recommendations based on their stated preferences. The system was developed using the Python programming language on the Google Colab platform and utilized a dataset consisting of 100 motorcycle spare part products with attributes including product name, motorcycle compatibility, product category, price, and rating. The recommendation process was carried out by matching user preferences with product attributes, after which each product was scored using a weighted calculation with compatibility weighted at 45%, product category at 30%, price at 15%, and rating at 10%. The system was evaluated using 10 testing scenarios by assessing the Top-1 recommendation. The evaluation results showed that all testing scenarios successfully generated recommendations that matched user requirements, achieving an accuracy of 100%. The findings indicate that the implementation of the Knowledge-Based Recommendation method, combined with a weighted attribute mechanism based on motorcycle compatibility, product category, price, and rating, is capable of producing recommendations that align with user preferences without requiring users' purchase history or rating history. Furthermore, the proposed method was found to be effective for motorcycle spare parts recommendation systems and has the potential to assist users in selecting appropriate products more quickly and accurately.

Implementasi Algoritma Decision Tree dan Ensemble Learning Berdasarkan Spesifikasiperangkat Keras Untuk Klasifikasi Harga Laptop

2026-07-27T10:02:08+07:00

The complexity of hardware specifications in laptops often makes it difficult for consumers to estimate price suitability in the digital market. This study aims to implement a laptop price classification system by comparing the performance of the baseline Decision Tree algorithm and the Ensemble Learning method (Random Forest) utilizing the Knowledge Discovery in Databases (KDD) approach. Contrary to the initial hypothesis that the ensemble model would provide significant performance improvements, the evaluation results revealed a paradoxical finding. Both classification models produced identically exact performance with an accuracy rate of 73.39% and an F1-Score of 73.41%. Technical analysis indicates that this anomaly is caused by the narrow dimension of deterministic features in the dataset, where RAM capacity and CPU architecture attributes dictate the decision boundaries absolutely, rendering the addition of hundreds of decision trees in the Random Forest computationally redundant. The results of this study contribute theoretical insights regarding the efficiency limitations of ensemble algorithms on low-dimensional datasets, while proving that a single Decision Tree is optimal and computationally efficient enough to be implemented as an inference engine in laptop price recommendation systems.

Rancang Bangun Sistem Informasi Pengaduan Fasilitas Publik pada Dinas PUPR Kabupaten Bima Menggunakan Prototyping dan Black Box Testing

2026-07-30T21:06:16+07:00

The management of complaints regarding public facility damage at the Public Works and Spatial Planning Agency (Dinas PUPR) of Bima Regency currently faces challenges, specifically regarding the lack of optimal integration across the processes of report submission, verification, disposition, follow-up, and monitoring. This situation results in an unstructured complaint management process and hinders the monitoring of report handling progress. This research aims to design and develop a web-based Public Facility Complaint Information System capable of supporting structured complaint management aligned with the workflow of the Bima Regency PUPR Agency. The system development method employed is Prototyping, which allows for the evaluation and refinement of the prototype based on user feedback. System design utilizes the Unified Modeling Language (UML), encompassing Use Case Diagrams, Activity Diagrams, Entity Relationship Diagrams (ERD), and user interface design. The system is implemented using the Laravel framework and a MySQL database. System testing was conducted using the Black Box Testing method across 12 test scenarios covering core system functions—ranging from user management, complaint submission, verification, disposition, follow-up, and monitoring to report generation. Test results indicate that all 12 scenarios were executed successfully, yielding the expected outcomes with a 100% success rate. The contribution of this research is the creation of a complaint information system that integrates report submission via Village Operators, verification and disposition by the General Admin, follow-up by Division Admins, and monitoring by the Head of the PUPR Agency. Consequently, the designed and developed system facilitates a more structured, well-documented, effective, and easily monitored process for managing public facility complaints.

Implementasi Arsitektur Hybrid CodeIgniter 4 dan Python dengan Enkripsi AES pada Sistem Smart Counseling Berbasis Web

2026-07-30T21:36:57+07:00

Counseling services at MAN 2 Garut are still dominated by conventional approaches that limit students’ privacy, making it difficult for them to openly express emotional concerns. This study aims to design and develop a web-based Smart Counseling system that supports counseling services in a safer, more responsive, and structured manner through a hybrid architecture. The development method used is Agile Scrum with the integration of the CodeIgniter 4 framework as the user interface and a Python backend as a microservice for data security. The system applies the Advanced Encryption Standard (AES) 256-bit algorithm to secure message data and a chatbot feature based on rule-based Natural Language Processing (NLP) to provide initial responses to users. Functional testing using the Black Box method on 11 scenarios showed that all main features worked as expected, including user authentication, mood tracking, counseling scheduling, chat encryption, two-way communication between students and the school counselor, and user account management by the admin. The results indicate that the developed system can support counseling services that are more inclusive, secure, and accessible for students. The main contribution of this study is integrating AES-based data security, mood-tracking emotion monitoring, and hybrid architecture into a single unified Smart Counseling system.

Prediksi Cuaca Harian Menggunakan Algoritma Long Short-Term Memory (LSTM) Berdasarkan Data Meteorologi Tahun 2025

2026-07-30T21:46:52+07:00

Daily weather prediction is crucial in supporting decision-making in the agriculture, transportation, disaster mitigation, and community activities influenced by atmospheric conditions. The weather in Bima City is dynamic, requiring a predictive model capable of learning sequential meteorological data patterns. This study aims to build a daily average temperature prediction model using the Long Short-Term Memory (LSTM) algorithm based on 2025 meteorological data. The contribution of this study is to develop an LSTM-based daily average temperature prediction model using multivariate meteorological data from Bima City combined with pre-processing steps in the form of missing value handling using moving averages, MinMaxScaler normalization, and time series data formation using a 30-day sliding window. This study also provides an initial evaluation of the application of LSTM to local meteorological data from Bima City, which has been studied only limitedly, as a basis for developing a deep learning-based weather prediction system. The variables used include minimum temperature (TN), maximum temperature (TX), average temperature (TAVG), average air humidity (RH_AVG), rainfall (RR), sunshine duration (SS), and average wind speed (FF_AVG). The test results show that the model produces a Root Mean Square Error (RMSE) value of 0.7242 and is able to follow the daily temperature change pattern in the actual data. The prediction results on the test data also show that most of the predicted values have a relatively small difference compared to the actual values, so the model is able to describe the daily temperature change pattern quite well. Based on the predicted weather parameters, the model is able to provide information about daily weather conditions, namely sunny, cloudy, and rainy, according to the values of rainfall, air humidity, and sunshine duration produced. This predicted information is expected to help the community as an initial picture of future weather conditions so that it can support the planning of various daily activities. However, the results of this study are still limited to one prediction method and have not been compared with other methods. Therefore, further research can conduct comparisons with other algorithms to improve the accuracy of weather predictions in Bima City.

Implementasi One-Dimensional Convolutional Neural Network untuk Klasifikasi Kualitas Air Depot Reverse Osmosis secara Real-Time berbasis Internet of Things

2026-08-02T13:16:26+07:00

Water quality monitoring in refill reverse osmosis (RO)-based drinking water depots in Indonesia requires continuous surveillance because membrane performance degradation due to fouling can occur gradually without visual indication, potentially endangering consumer health. This study aims to implement and evaluate a One-Dimensional Convolutional Neural Network (1D-CNN) architecture for real-time water quality classification on an IoT-equipped RO filtration system at Politeknik Negeri Manado. The system simultaneously reads eight sensor parameters at two measurement points (inlet and outlet), comprising turbidity, pH, Total Dissolved Solids (TDS), and temperature. A synthetic dataset of 5,000 samples representing operational condition variations is classified into three classes (Normal, Warning, Danger) based on thresholds from Indonesian Ministry of Health Regulation No. 2/2023. Preprocessing includes Min-Max Scaling and a sliding window technique (size 10 timesteps) to construct three-dimensional input tensors. A compact 1D-CNN model with only ~3,000 parameters is trained using the Adam optimizer with early stopping to prevent overfitting. The main contributions include a Dual-Protection mechanism integrating deterministic regulation-based rules with CNN inference, and an Explainability Engine generating textual diagnostics in Indonesian for field operators. Evaluation results demonstrate 99.80% accuracy with only 1 misclassifications from 500 actual test samples, proving the proposed approach effective for automated, accurate, and interpretable real-time water quality monitoring in RO depots.

Prediksi Risiko Penyakit Jantung Menggunakan Support Vector Machine dengan Seleksi Fitur dan Optimasi Hyperparamete

2026-08-09T12:16:29+07:00

Heart disease remains one of the leading causes of mortality worldwide, highlighting the need for accurate prediction models to support early diagnosis and clinical decision-making. This study proposes a heart disease risk prediction model based on Support Vector Machine (SVM) integrated with Recursive Feature Elimination (RFE) for feature selection and GridSearchCV for hyperparameter optimization. The study utilized the Cleveland Heart Disease Dataset, consisting of 303 patient records, 13 predictive attributes, and one target variable. The research workflow included dataset collection, data preprocessing, feature selection, hyperparameter optimization, model development, and performance evaluation using Accuracy, Precision, Recall, F1-score, and ROC-AUC. Experimental results demonstrate that the proposed model achieved an Accuracy of 86.89%, Precision of 85.71%, Recall of 85.71%, F1-score of 85.71%, and ROC-AUC of 93.18%. Feature selection successfully reduced irrelevant attributes, improving model efficiency, while hyperparameter optimization produced a more effective parameter configuration than the default settings. These findings indicate that integrating RFE with GridSearchCV enhances the predictive performance of SVM and provides a promising approach for supporting heart disease diagnosis using machine learning techniques.