TIN: Terapan Informatika Nusantara

Strategi Penanganan Pengecualian Pasif, Reaktif, Antisipatif: Simulasi Diskrit Ketahanan Operasional Layanan EEG Rawat Jalan

2026-10-31T14:14:25+07:00

Outpatient electroencephalography (EEG) services face service-time variability, resource constraints, and operational disruptions that increase patient waiting time and Length of Stay (LoS). This study develops a High-Fidelity Discrete-Event Simulation (HF-DES) model that explicitly represents disruptions as operational exceptions in order to compare three handling strategies, namely Baseline (S1, passive), Active Handling (S2, reactive), and Anticipatory Resilience (S3, anticipatory), in terms of both efficiency and operational resilience. The model integrates service flows, probability distributions derived from Quantile Matching Optimization, resource capacities, operational calendars, and disruption mechanisms, and was verified and validated using the Mean Absolute Percentage Error (MAPE) and statistical tests. Experiments used 30 replications per scenario with Common Random Numbers and Holm–Bonferroni correction, together with stress testing through increased demand, machine disruptions, and service durations. The model is highly accurate (MAPE 0.59%; p > 0.05). Scenario S3 performs best, with a LoS of 63.59 minutes (75.73% lower than S1), a waiting time of 6.17 minutes, and the lowest Robustness Index of 12.7% (S1: 571.7%; S2: 16.8%). The superiority of S3 is theoretically grounded: because administrative completeness and resource readiness are verified before patients enter the critical pathway, exceptions are prevented from propagating into the queue, so process variability and cascading delays are reduced at their source rather than recovered after they occur; in queueing theory this reduces waiting time most strongly when utilization is high. Managerially, hospitals can improve both the efficiency and the resilience of EEG services through procedural redesign and hospital information system integration, without adding physicians, technicians, or machines.

Optimizing Team Formation Using Enhanced Snake Optimizer Algorithm (ESOA): A Comparative Study with Metaheuristic Approaches

2026-10-31T14:14:26+07:00

Team formation (TF) is an important problem in various domains because it involves selecting people with complementary skills while minimizing the cost of communication. Since metaheuristic algorithms effectively handle complex optimization problems. This study proposes the Enhanced Snake Optimizer Algorithm (ESOA) to solve the TF problem and evaluates its effectiveness against the Snake Optimizer Algorithm (SOA), Jaya Algorithm (JA), and Sine–Cosine Algorithm (SCA). ESOA strengthens SOA using opposition-based learning, dynamic parameter adjustment, sine–cosine perturbation, and chaotic/Cauchy mutation to reduce premature convergence and local-optimum stagnation. Experiments were conducted on a real TF dataset from Universiti Malaysia Pahang Al-Sultan Abdullah with required-skill-set sizes of 10, 20, 30, 40, and 50, using 10 independent runs per setting. Performance was assessed using team size (TS), team cost (TC), and runtime. ESOA achieved the lowest best TC; for 50 required skills, ESOA obtained best TC = 402.641 compared with SOA = 408.026, JA = 413.854, and SCA = 425.051, while maintaining comparable TS. Although ESOA incurred higher runtime, the results indicate improved solution quality and robustness for complex TF instances, supporting practical team assembly in collaborative environments.

Penerapan Metode Reorder Point pada Sistem Kasir dalam Menentukan Waktu Pengadaan Barang

2026-10-31T14:14:26+07:00

Toko Kardus Mitra Packing still manages transactions and inventory conventionally, resulting in obstacles such as the lack of real-time stock deduction, slow cashier services, and the determination of procurement times based solely on subjective estimates, which triggers an average stockout frequency of 5 to 8 times per month for primary packaging products and a 42% cashier service delay during peak hours. If not addressed immediately, this condition will result in the loss of sales potential and disrupt logistics availability, especially during order surges from customers. The purpose of this study is to develop a mobile and web-based cashier system by implementing the Reorder Point (ROP) method as a solution for automatically determining procurement times. The primary contribution and novelty of this research lie in the real-time architectural integration between a front-line mobile Point of Sale application and a server-side probabilistic Reorder Point computation engine, enabling every sales transaction to dynamically recalculate daily demand standard deviation and safety stock thresholds without separate manual analysis. This system was built using the Flutter framework for employee operations, Laravel for the owner's dashboard, and a MySQL database. Functionality testing was conducted using the Black Box Testing method. The test results showed that the system functioned with a 100% success rate in all test scenarios and reduced transaction processing time by 68.5% to an average of 55 seconds per transaction. The final result of this research is a cashier system capable of automating daily transactions, synchronizing inventory in real-time, and mathematically calculating safe stock limits to provide early procurement warnings on the dashboard, thereby significantly facilitating the store owner in making more accurate and data-driven restocking decisions.

Analisis Komparatif C4.5, XGBoost, dan CatBoost Berbasis SMOTE untuk Prediksi Kelayakan Calon Karyawan

2026-10-31T14:29:43+07:00

Recruitment is an important stage in helping organizations identify candidates who meet job requirements. Manual assessment of a large number of candidates can be time-consuming and may lead to inconsistent decisions. This study aims to compare the performance of C4.5, XGBoost, and CatBoost in classifying candidate eligibility and to determine the model with the most consistent performance. The AI Fair Recruitment Dataset was used, consisting of 119,971 candidate records. The study followed the CRISP-DM framework, including Business Understanding, Data Understanding, Data Preparation, Modeling, Evaluation, and Deployment. Data preparation included attribute selection, encoding, a stratified split of 90% training data and 10% test data, and class imbalance handling using SMOTE applied only to the training data. The models were compared using accuracy, precision, recall, F1-score, ROC-AUC, and confusion matrix. The results show that XGBoost and CatBoost achieved 98% accuracy, while C4.5 achieved 95%. CatBoost achieved 96% precision, 98% recall, 98% F1-score, and 97% ROC-AUC, providing the most consistent performance among the three models. The confusion matrix also shows that CatBoost produced the lowest classification errors, with 121 False Positive and 50 False Negative cases. These results indicate that CatBoost was the most suitable model under the experimental setting of this study. The model was subsequently implemented in a Flask-based prototype as an initial screening support tool. This study did not quantitatively evaluate algorithmic fairness; therefore, high predictive performance should not be interpreted as evidence that the model is bias-free.

Sistem Deteksi dan Monitoring Terpusat Penyakit Daun Padi Berbasis Mobile Menggunakan MobileNetV2

2026-10-31T14:39:59+07:00

Rice leaf diseases such as Bacterial Leaf Blight, Brown Spot, and Leaf Smut are a major threat to agricultural productivity in Indonesia because manual identification by farmers remains subjective, slow, and error prone. This study aims to develop a rice leaf disease detection system that integrates a Flutter based mobile application with a web based monitoring dashboard, so that diagnosis can be performed independently by farmers while being monitored centrally by system administrators. The method applied is a Convolutional Neural Network with a transfer learning approach on the MobileNetV2 architecture, trained using an expanded dataset of 300 rice leaf images obtained from primary field observation data in Klaten and secondary data from the Kaggle platform, then converted to TensorFlow Lite format to enable on-device inference on mobile devices. The system backend was built using Flask and a MySQL database to manage user data and detection history. Experiments on four hyperparameter scenarios showed that a configuration of 20 epochs, batch size 32, and learning rate 0.001 produced the best performance with a test accuracy of 95.24 percent and a loss of 0.1506 without significant indication of overfitting. The final model size of 9.71 MB with an average inference time of under 2 seconds proves that this system is feasible to be applied as an accurate, lightweight, and responsive rice disease diagnosis tool in the field.

Klasterisasi Pola Seismik: Perbandingan K-Means dan ST-DBSCAN pada Data Gempa Spasio-Temporal

2026-10-31T14:53:50+07:00

Earthquakes are natural disasters with significant socio-economic impacts; thus, understanding their distribution patterns and characteristics is essential for disaster mitigation. Indonesia is situated along the Pacific Ring of Fire with intense seismic activity, and the volume of earthquake data recorded by BMKG is too vast for manual analysis. This study compares K-Means and ST-DBSCAN using 93,815 spatio-temporal data points from a historical catalog (2008–2023) and data from the final 30 days of 2026. The number of K-Means clusters was determined using the elbow method supported by the Silhouette Score and Davies-Bouldin Index (DBI); both algorithms were evaluated using the Silhouette Score, DBI, and Adjusted Rand Index (ARI) across six tectonic zones. K-Means (k = 4) outperformed ST-DBSCAN across all three metrics (Silhouette Score: 0.1563, DBI: 2.7180, ARI: 0.4137), whereas ST-DBSCAN produced 20 clusters with a single cluster containing 70.9% of the data. Three K-Means clusters corresponded to the Sulawesi–Maluku, Sunda Arc, and Sumatra regions, while one represented a temporal artifact from the time_days feature. An ablation test excluding time_days (k = 3) improved the ARI to 0.5102. Results are presented via the Flask-based NusaSeis web dashboard with interactive maps and automated interpretations by Gemini 2.5 Flash; all 23 black box scenarios were validated. NusaSeis serves as an open seismic exploration platform enabling academics and disaster management agencies to interactively access national seismicity summaries without programming expertise, though it is not yet suitable for official tectonic zoning or hazard assessment.

Pemanfaatan Model Context Protocol untuk Otomatisasi Root Cause Analysis pada Microservices Berbasis Docker

2026-10-31T15:06:19+07:00

Microservices architecture deployed in Docker container environments offers high scalability, yet generates large and segmented telemetry data that prevents conventional monitoring methods from supporting rapid root cause analysis. Administrators must manually trace hundreds of metrics and logs through complex dashboards, while existing chatbot approaches remain rule-based and limited to presenting raw data without deep analysis. This study designs and implements a monitoring system that integrates a Large Language Model (LLM) through the Model Context Protocol (MCP) to automate disruption analysis on Docker-based microservices. A custom MCP Server was developed by modifying the open-source Prometheus-MCP-Server repository to bridge two telemetry sources: container metrics collected by Prometheus via cAdvisor, and container logs retrieved from the Docker Daemon via the Docker SDK, combined with a natural-language interface provided by the LLM through Cursor as the MCP Client. The research object consisted of eight e-commerce microservice containers running on a Virtual Private Server. Telemetry accuracy validation yielded a mean MAE of 0.1071 MB and a mean MAPE of 0.29% across all containers. Response-time testing showed the system responding within 3-23 seconds, with the LLM reasoning component dominating 98.72%-99.81% of total response time, while the MCP Server contributed only 0.044 seconds of overhead. Scalability testing with 1-10 concurrent users showed elastic CPU usage and stable RAM consumption within 165.7-169.3 MB. These findings demonstrate the technical feasibility of MCP-based LLM integration for supporting microservices incident analysis through unified access to container metrics and logs, while highlighting limitations in RCA diagnostic accuracy evaluation and production-scale scalability testing.

Tren Teknologi dan Mikrokontroler pada Sistem Pakan Ikan Otomatis Berbasis IoT: Tinjauan Sistematis 2022–2026

2026-10-31T15:16:14+07:00

This study aims to analyze technology trends and microcontrollers in automatic fish feeding systems based on the Internet of Things (IoT) through a Systematic Literature Review (SLR) of publications from 2022–2026. The method used is SLR with the PRISMA 2020 protocol, bibliometric analysis assisted by VOSviewer, and content analysis. Data were taken from the ScienceDirect database, with an initial search yielding 200 articles for Research Question 1 (RQ1) and 50 articles for RQ2. After identification, screening, and eligibility assessment, 32 articles for RQ1 and 17 articles for RQ2 were included for further analysis. The results show a shift from simple timer-based systems toward intelligent architectures integrated with AI and machine learning, with IoT+AI dominating at 35%. The Arduino platform dominates at 60%, followed by ESP32/NodeMCU, which is increasing due to wireless connectivity. The most critical sensors are temperature (30%), pH (25%), and dissolved oxygen (25%), while feeder actuators dominate at 30%. Publications peaked in 2025 with 26 articles. For RQ2, Arduino family microcontrollers (Uno, Mega, Nano) are the most dominant, covering two-thirds of articles, but there is a gradual shift toward ESP32 and ESP8266. An important finding reveals a reporting gap where 70% of articles do not explicitly list hardware specifications, hindering research replication and technology adoption. This study provides a current technology trend map and evidence-based recommendations for researchers and farmers in selecting the most mature technology for sustainable automatic feeding systems in the aquaculture sector.

Perancangan Sistem Manajemen Proyek Internal Menggunakan Design Science Research Methodology pada Perusahaan Infrastruktur dan Teknologi

2026-10-31T15:23:55+07:00

Infrastructure and technology companies still face challenges in project management because many operational activities are manual, fragmented, and not yet integrated, causing project data management, budget planning, progress monitoring, and reporting to operate inefficiently and delay accurate decision-making. This study aims to design and develop a web-based internal project management system that integrates multi-company master data management, budget planning with multi-addendum support, daily and weekly progress monitoring with S-curve visualization, analytical dashboards, automated PDF reporting, and information distribution through a Telegram Bot. The study applies Design Science Research Methodology (DSRM), which consists of six stages: problem identification and motivation, definition of solution objectives, design and development, demonstration, evaluation, and communication. The system is implemented using Vite, TypeScript, and Tailwind CSS for the frontend, Node.js and Express.js for the backend, and PostgreSQL as the relational database system. Evaluation is conducted through User Acceptance Testing (UAT) involving 29 functional test scenarios across eight module categories, and all categories achieved a 100% success rate. These results indicate that the proposed system improves project monitoring efficiency, maintains reporting consistency, accelerates information delivery to stakeholders, and supports the digitalization of internal project management in an integrated manner. The contribution of this study is the development of an integrated project management system artifact that combines budget management, progress monitoring, project analytics, and information distribution within a single web-based platform to support more structured and measurable internal project management.

Perancangan Sistem Booking Lapangan Olahraga Basket Berbasis Web Menggunakan Metode Research and Development (R&D) dalam Meningkatkan Efisiensi Pengelolaan Lapangan

2026-10-31T15:37:14+07:00

Basketball court reservations are still predominantly made through conventional methods, such as phone calls, in-person visits, and private messages, which result in scheduling conflicts, double bookings, and inefficient administrative processes. This study aims to develop a web-based basketball court reservation information system to improve the efficiency of court management. This study uses the Research and Development (R&D) method with the Waterfall model, which includes requirements analysis, system design, implementation, testing, and maintenance. Data was collected through observation, structured interviews, and questionnaires. The system was developed using PHP, MySQL, HTML, CSS, and JavaScript, and is equipped with features for real-time schedule management, online booking, payment confirmation, and automated transaction reports. The results demonstrate that the website has high access speed, is responsive, and is stable for user use. Functional testing using the Black-Box Testing method showed that all system features operate as expected. After the Black-Box testing was completed, the web system was also tested for overall system speed. Based on Google PageSpeed Insights/Lighthouse testing, the web-based booking system demonstrated excellent and efficient performance across various devices. On mobile, the system achieved a performance score of 88%, an accessibility score of 93%, and a best practices score of 96%. On desktop, the system achieved a perfect performance score of 100%, with an accessibility score of 93% and a best practices score of 96%. Based on the test results, the web-based booking system was deemed highly suitable and received very positive feedback from both subject matter experts and end users. The subject matter experts’ validation of 20 indicators across eight aspects yielded an average score of 77 out of a maximum of 80, with a suitability percentage of 96.25%, which falls into the “Highly Suitable” category. Meanwhile, the results of the user response test, which involved managers and tenants, yielded an average score of 76.67 out of 80, with a 95.83% achievement rate, placing it in the “Highly Feasible” category. These results indicate that the system not only meets the criteria for technical and content validity but is also capable of providing benefits and effectively supporting operational activities and on-site booking processes.

Simulasi Sistem Kendali Daya ON/OFF Komputer Desktop Melalui Jaringan Internet Berbasis ESP32

2026-10-31T15:58:01+07:00

Desktop computers are well known for their high flexibility and are widely used for advanced computing tasks such as software development, graphic design, and other resource-intensive applications. However, they remain limited in portability due to their relatively large physical size. To overcome this, desktop computers can be accessed remotely over the Internet using software such as Moonlight and Parsec. Nevertheless, this approach introduces another issue, as the computer must remain powered even when not in use, resulting in unnecessary energy consumption. This study aims to develop an Internet of Things (IoT)-based system capable of remotely controlling the power state of a desktop computer, turning it on and off as needed. The research adopts a systems engineering approach using the prototyping method, carried out iteratively from requirements analysis to testing and evaluation. The system utilizes an ESP32, Firebase Realtime Database, and a Flutter mobile app as the user interface. The results demonstrate reliable inter-device communication and remote computer power control with total average response time of approximately 50ms. The device has a compact form factor and is self-powered using the computer's 5V standby power supply. The Flutter mobile app exhibits a responsive user interface, with a render raster time of approximately 2ms on a 60 fps refresh rate. The developed system successfully enables remote computer power control, improving access flexibility while reducing unnecessary energy consumption, directly contributing to the advancement of science and technology, particularly in the application of IoT.

Efektivitas Penggabungan Data Lintas Provinsi dan Augmentasi TimeGAN pada Prediksi Harga Beras Berbasis LSTM

2026-11-04T10:28:37+07:00

Rice price fluctuations in each province are difficult to predict, although accurate forecasts are needed to anticipate price shocks, while the multivariate data available for several provinces are often short. This study compares two ways of addressing this limitation in a Long Short-Term Memory (LSTM) model: pooling data from three provinces (East Java, Central Java, and West Java) into a single model, and adding synthetic data from a Time-series Generative Adversarial Network (TimeGAN). Monthly data (March 2024 to March 2026) from four sources (weather, rice prices, food reserve stocks, and exchange rates) yielded 25 observations per province. The 84 scenarios under three training conditions (no augmentation, 50%, and 100% augmentation) made 252 experiments, assessed with MAPE, RMSE, and MAE. The pooled model K7 was more accurate than local models in all three provinces. Without augmentation, mean MAPE fell from 2.056% to 1.258%, RMSE from IDR 326.777/kg to IDR 223.018/kg, and MAE from IDR 306.063/kg to IDR 187.795/kg. In contrast, TimeGAN augmentation produced no detectable improvement, and at 100% it tended to raise the pooled models’ MAPE from 1.541% to 1.710%. The synthetic data were easily distinguishable from the real data (mean discriminative score 0.893). For food price monitoring in regions with short data, collecting real data from other provinces appears more promising than building synthetic data, so price-forecasting systems can prioritize pooling data across provinces before adding generative models.

Evaluasi Kritis Generalisasi Multidataset pada Deteksi Hoax Berbahasa Indonesia Menggunakan BiLSTM dan IndoBERT

2026-11-04T10:39:15+07:00

Machine learning-based Indonesian hoax detection often reports high scores on test data sharing the training data’s distribution, although such scores may misrepresent actual model capability. This critical evaluation, not a new method, analyzes how multidataset composition affects BiLSTM and IndoBERT performance and generalization, and identifies indications of shortcut learning arising from spurious correlations between labels and dataset origin. Three datasets were used: Mendeley (426 samples), Komdigi (3,951 samples), and sh1zuka (8,000 sampled), combined into 12,377 samples (8,256 Valid, 4,121 Hoax). Evaluation employed three schemes: Scheme 1 (stratified 70:15:15 split of the combined data), Scheme 2 (forward cross dataset), and Scheme 3 (reverse cross dataset). In Scheme 1, BiLSTM and IndoBERT achieved Macro F1 scores of 0.9803 and 0.9814, but these fugures cannot be read as successful hoax detection, since under cross dataset schemes Macro F1 dropped to 0.3642 and 0.5624 (Scheme 2) and 0.4231 and 0.4109 (Scheme 3), with accuracy below the majority class baseline. Dataset 1, comprising only 3.4% of the Scheme 1 test data, accounted for 87.9% of BiLSTM errors and 100% of IndoBERT errors, while the accuracy of the word “beredar” as a single label predictor fell from 96.30% on combined data to 42.96% on Dataset 1. These findings indicate that high scores under conventional evaluation reflect recognition of dataset origin more than comprehension of news content; therefore, cross dataset evaluation and source control tests should accompany random splitting on multisource data.

Habit Tracker Application Using the Gamification Method to Improve Student Discipline

2026-11-04T10:46:21+07:00

Student discipline in academic activities remains a continual challenge, commonly due to inadequate time management and frequent electronic distractions. Existing Habit Tracker applications typically focus on recording activities, with limited capacity to sustain user motivation. This study describes the design, development, and evaluation of an Android-based Habit Tracker application that incorporates seven gamification elements: points, levels, badges, streaks, leaderboards, rewards, and reminders. The application seeks to improve student discipline at Universitas Muhammadiyah Banten. This study contributes a Habit Tracker that integrates seven gamification elements in a single application aimed at academic discipline, together with empirical evidence on its user acceptance and on changes in student discipline scores. Employing a quantitative One Group Pretest-Posttest design and the Prototype development method, data were collected from 30 students across the 15th, 16th, and 17th cohorts using pre-test, post-test, and User Acceptance Test (UAT) questionnaires on a Likert scale. Data analysis included validity testing (Corrected Item-Total Correlation), reliability testing (Cronbach's Alpha), and a Paired Sample T-Test to compare discipline scores before and after application use. The UAT indicated a user acceptance rate of 75.37%, categorized as good. The Paired Sample T-Test demonstrated an increase in mean discipline scores from 31.53 to 38.27, with a t-value of -11.914 and a significance value of 0.000 (p<0.05), leading to the rejection of H0 and acceptance of H1. The results indicate that the gamification-based Habit Tracker application developed in this study is associated with a significant increase in student discipline scores at Universitas Muhammadiyah Banten.

Implementasi Bcrypt dan SHA-256 untuk Keamanan Autentikasi dan Verifikasi Pengguna Aplikasi Persediaan Barang

2026-11-04T11:07:46+07:00

User authentication and verification security are essential aspects of web-based applications, as insecure credential storage may increase the risk of password disclosure and unauthorized access to the system. This study aims to implement and evaluate a user security mechanism in an inventory application by employing bcrypt for password protection, SHA-256 for verification token generation, and One-Time Password (OTP) for email ownership verification. The study adopted a software engineering approach consisting of security requirement identification, security mechanism design, implementation, and functional and performance testing. Bcrypt was implemented with a cost factor of 12 in the application, while its performance was evaluated using cost factors of 8, 10, 12, and 14 by measuring password hashing and verification times. SHA-256 was employed to generate a verification token from a combination of the user's email address, time information, and a unique identifier, whereas a six-digit OTP was delivered via email and validated against the value temporarily stored in the session. The functional testing results demonstrated that the system successfully authenticated verified accounts with correct passwords while rejecting incorrect passwords, unverified accounts, and invalid OTPs. Increasing the bcrypt cost factor resulted in higher computational time. The hashing times for cost factors 8, 10, 12, and 14 were 45.2 ms, 98.7 ms, 385.4 ms, and 1,542.8 ms, respectively, while the corresponding verification times were 43.8 ms, 96.3 ms, 382.1 ms, and 1,538.6 ms. In the email verification process, OTP generation required 2.1 ms, email delivery required 847.3 ms, and OTP validation required 1.8 ms. These results indicate that bcrypt, SHA-256, and OTP can be integrated as distinct yet complementary security mechanisms, while the selection of an appropriate bcrypt cost factor should consider the trade-off between security requirements and system processing time.

Rancang Bangun Purwarupa Sistem Pewarisan Aset Digital Berbasis Dead Man's Switch Dengan Enkripsi AES-256

2026-11-04T11:16:37+07:00

Digital assets may become inaccessible to heirs when their owners die or experience emergencies. This study develops a web- and mobile-based digital inheritance prototype integrating a Dead Man’s Switch (DMS), server-side AES-256 encryption, and folder-based access control. Development followed the Waterfall model and Unified Modeling Language (UML), using Laravel for the RESTful API backend, React Native for the interface, MySQL as the database, and a cron job to check inactivity every minute. Users perform periodic check-ins; once the time limit expires, the account status changes to released and access is processed according to heirs’ permissions. Email notifications contain a download link valid for three days and a One-Time Password (OTP) for heir authentication. Black-box testing of eight functional scenarios showed that all scenarios operated as expected. Across 25 repetitions of the access-release scenario, emails arrived 4–12 seconds after the time limit was reached. Testing was limited to functionality and notification timing; load testing, penetration testing, and user evaluation were not conducted. The prototype integrates activity monitoring, data protection, and controlled cross-platform access, but does not replace a legally valid will. Further security validation and user evaluation are needed before the system is used to manage high-value digital assets.

Aplikasi Pengingat Gamifikasi dengan Notifikasi Lokal Offline-First dan Validasi Poin Server-Side: Evaluasi Skalabilitas dan Kebergunaan

2026-11-04T12:01:59+07:00

Procrastination is a barrier to the productivity of students and workers in the digital era, while conventional reminder applications tend to be passive and may experience notification delays due to push notification latency and mobile operating system power optimization policies. This study aims to design, implement, and evaluate DoYourTask, a hybrid client-server mobile reminder application integrating offline-first local notifications using exactAllowWhileIdle with gamification features, including experience points (XP), levels, badges, and leaderboards, supported by centralized server-side validation. Its motivational foundation draws on Self-Determination Theory (SDT), particularly the needs for competence and autonomy. Evaluation comprised Black Box functional testing using equivalence partitioning and boundary value analysis, RESTful API load testing with Apache Benchmark at 2, 5, and 10 virtual users (VU), MySQL query scalability testing with up to 10,000 task records, and a usability pilot study involving 10 respondents using the System Usability Scale (SUS). The results showed that all 21 functional test scenarios passed, yielding a 100% success rate. API testing achieved a 100% transaction success rate, a 0.00% error rate, and throughput of 2.75–3.29 requests per second, while median latency increased from 601 ms to 3,669 ms. The idx_users_xp composite index reduced inspected rows by 99% and improved leaderboard query performance by 26.9%. The mean SUS score was 80.5 ± 9.2, corresponding to Grade B and an Excellent rating. The results demonstrate the technical feasibility of the hybrid architecture and indicate good usability based on an initial evaluation.

Integrasi Large Language Model dengan FAQ Retrieval dan Sinkronisasi Basis Pengetahuan Berbasis Web Scraping pada Chatbot Layanan Admisi

2026-11-04T12:06:32+07:00

University admissions services require timely and consistent answers, whereas manually maintained chatbot knowledge bases can quickly become outdated. This study develops an admissions chatbot integrating a Large Language Model, semantic FAQ retrieval, and knowledge base synchronization through web scraping. The system combines a Next.js interface, a Flask backend, Firebase Firestore, BeautifulSoup for static pages, Selenium for dynamic pages, and a generative model constrained by retrieved FAQ context. Development follows Design Science Research. Retrieval represents each FAQ as a 384-dimensional multilingual embedding, measures similarity using cosine similarity, and passes the top five FAQs to the model as answer context. Semantic retrieval is compared with a TF-IDF baseline, while answer quality is compared with an LLM-only configuration. Because operational data were unavailable, the evaluation used a controlled synthetic benchmark comprising 180 questions and 30 test URLs. In the simulation, semantic retrieval achieved 88.1% Hit@1, 93.8% Hit@3, and 0.909 MRR. FAQ-grounded generation improved simulated answer correctness from 72.2% for the LLM-only configuration to 91.1% and reduced unsupported answers from 17.8% to 5.0%. Direct functional testing in the deployed environment showed that four of seven scenarios passed. These findings indicate design feasibility, but field validation with real users remains necessary. The contribution of this study is an auditable orchestration design that connects official-source scraping, administrator curation, semantic retrieval, and the LLM in a single workflow, together with a separated evaluation framework for retrieval, answer quality, scraped-data quality, and system performance.

Implementasi Contextual Multi-Armed Bandit dengan Algoritma LinUCB untuk Rekomendasi Template pada Sistem Kampanye WhatsApp Berbasis Web

2026-11-04T12:11:33+07:00

Uniform WhatsApp campaign templates may overlook differences in timing and audience characteristics. This study develops an adaptive and auditable template recommendation module for a web-based campaign system. Design Science Research Methodology guides problem identification, objective definition, artifact development, demonstration, and evaluation. The disjoint Linear Upper Confidence Bound (LinUCB) algorithm treats Informative, Persuasive, and Urgency templates as three available actions. A 19-dimensional context vector encodes day, time bucket, audience, and a bias term. Template selection maximizes a linear reward estimate plus an exploration bonus, while the selected action is updated using delivered, read, and replied rewards weighted at 0.2, 0.3, and 0.5. Synthetic evaluation comprises 30 trials of 100 contexts each, with an exploration parameter of 0.3 and parent seed 1790543727863. LinUCB achieves a mean total reward of 39.520 with a standard deviation of 0.631, compared with 36.800 and 1.790 for the context-blind baseline. The mean improvement is 7.39%, with a mean regret of 2.769. The contribution is an integrated context–decision–feedback workflow with decision logging, operator control, and repeated evaluation isolated from operational data. These findings support artifact feasibility in the synthetic environment; improvements in real recipient responses require field validation.

Implementasi Sistem WhatsApp Blast Berbasis Web dengan Fitur Segmentasi Calon Mahasiswa Menggunakan K-Means Clustering

2026-11-04T14:43:28+07:00

The delivery of prospective student admission information through WhatsApp requires structured contact management and target-group selection. This study develops a web-based WhatsApp Blast system with prospective student segmentation using K-Means Clustering. The system uses React and Next.js, Node.js and Express, PostgreSQL, a Python preprocessing and clustering service, and REST APIs. The clustering variables are study-program interest, school of origin, school major, and domicile; names and WhatsApp numbers are retained only as identity attributes. Actual testing used 120 controlled synthetic records. The system stored 113 valid records, detected 3 duplicates and 4 invalid records, and recorded one missing value in each clustering attribute. All missing values were imputed as Tidak Diketahui, so all 113 valid records entered clustering with no exclusion. ILKOM was normalized to Ilmu Komunikasi, and One-Hot Encoding produced 35 features. Evaluation for K=2–6 selected K=5 through agreement among Elbow, Silhouette, and Davies-Bouldin. The K-Means run produced an SSE of 300.1250, a Silhouette Score of 0.1110, a Davies-Bouldin Index of 2.6207, and cluster sizes of 30, 27, 20, 19, and 17. All thirteen executed Black Box scenarios passed. These results validate the implementation on synthetic data and do not generalize the prospective-student population.

Chatbot Admisi Berbasis LLM dengan Retrieval-Augmented Generation, Ekstraksi Data Prospek, dan Dashboard Analitik

2026-11-04T15:07:38+07:00

New-student admission services face high inquiry volumes during enrollment periods and the limitations of rule-based chatbots in handling natural-language variation. This study develops and audits an admission chatbot based on a large language model that augments each request with the entire Firestore FAQ collection (non-selective full-context augmentation rather than ranked retrieval), supported by prospect-data capture, user feedback, an analytics dashboard, and Gemini-assisted knowledge-base scraping. The audit covers the Flask backend, test files, the production Next.js frontend bundle, and read-only Firestore aggregations that do not expose personal data. A 28 September 2026 snapshot recorded 238 FAQs, 191 chat logs, 9 leads, and 14 feedback records; 9 were helpful and 5 unhelpful, yielding a 64.3% helpfulness rate. The earlier 8/8 result is retained only as a test-dataset result rather than the production knowledge-base size. The average Heuristic Lexical Relevance Score of 0.54 and the Lexical Grounding Indicator range of 0.27–0.80 were derived only from illustrative text and are not treated as validated semantic or hallucination-detection measures. An internal Cache Microbenchmark recorded a 99.2% reduction in serialization overhead rather than end-to-end latency, while the previously reported 90% cache hit rate could not be reproduced. The initial audit reported a suspected static administrative token in the frontend bundle; the developer stated that dashboard access uses Firebase Authentication for registered accounts. Token classification and backend authorization still require code verification. Additional public-interface testing produced relevant responses in four UPJ-related cases and refusals in two out-of-domain cases, limited to six purposively selected cases. Prospect conversion and CSAT cannot yet be calculated because the funnel denominator and a CSAT instrument are unavailable. The main contribution is an implementation-grounded framework that separates production observations, test-dataset results, heuristic proxies, and outcomes still requiring experiments.

Analisis Performa Metode BiLSTM untuk POS Tagging pada Kalimat Bahasa Indonesia Tingkat Dasar

2026-11-04T15:16:27+07:00

The effect of hyperparameters on part-of-speech tagging (POS Tagging) performance for the Indonesian language in deep learning models is rarely examined in isolation, because earlier studies usually add complex components (pretrained embeddings, CNN, CRF, or BERT), so the effect of hyperparameters on a plain Bidirectional Long Short-Term Memory (BiLSTM) cannot be isolated. The novelty of this study is a stepwise analysis of five hyperparameters (embedding dimension, hidden dimension, learning rate, train-test split ratio, and number of epochs) on a three-layer BiLSTM built from scratch in NumPy, without pretrained embeddings or CRF, on basic-level Indonesian sentences, that is, sentences commensurate with the comprehension level of elementary school (SD) students, represented by first-grade SD/MI textbooks. Secondary data comprised 10,747 sentences with 23 labels, combining the IDN Tagged Corpus and 747 textbook sentences. The effect of hyperparameters proved highly uneven: learning rate and hidden-layer capacity determined performance, whereas embedding dimension, split ratio, and number of epochs shifted accuracy by less than 0.5 percentage points. The best configuration reached 95.92% accuracy and 0.9966 weighted ROC-AUC, with labels X, UH, and FW being the hardest to predict. A five-seed test of the final configuration showed stable results (accuracy 95.82 ± 0.08% at 50 epochs and 95.76 ± 0.17% at 98 epochs; the difference between the two was not significant, p = 0.26), but small differences between hyperparameter settings have not been tested statistically. This study provides a simple and transparent baseline and a priority order for hyperparameter tuning as a starting point for NLP practitioners.

Analisis Efektivitas Pembobotan TF-IDF dan Metode Cosine Similarity Untuk Deteksi Plagiarisme

2026-11-04T15:56:13+07:00

Plagiarism in academic documents undermines scientific integrity and significantly degrades the quality of education. This study aims to analyze the effectiveness of combining Term Frequency-Inverse Document Frequency (TF-IDF) statistical weighting with the Cosine Similarity distance metric for detecting the degree of global semantic similarity at the document level. The computational experiments were conducted using the Python programming language, supported by the Natural Language Toolkit (NLTK) and Scikit-Learn libraries. The test dataset consists of an internationally standardized secondary dataset from the PAN at CLEF 2011 corpus (a subset of the PAN-PC-2011 corpus), comprising 500 pairs of suspected (query) documents and their corresponding source documents. As the test dataset contains only positive pairs (instances of plagiarism) and no negative pairs (non-plagiarized instances), the binary classification evaluation focuses on True Positives (TP), False Negatives (FN), positive-class accuracy, and recall. The primary contribution of this research is the provision of empirical evidence regarding an optimal threshold (2%) and a comprehensive analysis of the impact of "vector dilution" in full-length documents. The highest recall achieved in the tested scenarios was 94% (470 TP out of 500 pairs) at a 2% threshold. Recall decreased as the threshold increased, reaching 0% at thresholds of 80% and 90%. These results indicate that the threshold significantly influences detection sensitivity for positive instances; however, it does not yet reflect the system's ability to correctly reject non-plagiarized documents. Consequently, this method is best positioned as an initial screening tool that requires further evaluation using both positive and negative pairs.

Integrasi Ensemble Learning dan Explainable AI untuk Prognostik Survival Lima Tahun Pasien Breast Carcinoma

2026-11-04T16:10:16+07:00

Five-year survival prediction in breast carcinoma cases has an essential role in supporting clinical decision-making. Nevertheless, limitations related to predictive capability and the interpretability of machine learning models continue to present challenges. In this research, a survival prediction framework was developed by combining ensemble learning methods with Explainable Artificial Intelligence (XAI). The study utilized secondary data from the SEER dataset obtained through Kaggle, covering 4,004 patient records. Several stages were conducted, including exploratory analysis, data preprocessing, and model construction using Random Forest, XGBoost, and CatBoost. These algorithms were subsequently integrated through a stacking mechanism in which a meta-learner was applied. Performance assessment involved accuracy, precision, recall, F1-score, and AUC-ROC measurements, while SHAP and LIME were employed to examine model interpretability. Experimental findings revealed that the stacking approach produced an accuracy of 0.6442, an F1-score of 0.7658, and a recall value of 0.8292, although it did not consistently outperform the standalone models. Among all approaches, CatBoost achieved the strongest predictive results with an accuracy of 0.6604 and an F1-score of 0.7901. In contrast, the stacking model recorded the highest AUC-ROC value of 0.5109 despite only a narrow difference. Further analysis demonstrated that tumor characteristics together with patients’ clinical conditions contributed most significantly to prediction outcomes. The integration of stacking-based ensemble learning and XAI in this study was intended to produce a predictive model that offers dependable performance while remaining more interpretable for clinical use.