Jurnal Sains dan Teknologi Informasi

Implementasi Kombinasi Pembobotan ROC Dengan Metode MAUT Dalam Sistem Pendukung Keputusan Pemilihan Security Patrol Terbaik

2026-10-01T10:15:57+07:00

Companies providing security services for BTS (Base Transceiver Station) tower sites have demonstrated strong performance to operators and BTS partners, evidenced by the high number of secure BTS towers under their protection. Security operations rely on a key initiative: the tower-to-tower site patrol program. The objectives of this activity include minimizing the risk of asset or equipment theft, preventing facility vandalism and site access blockages, and resolving site-related disturbances through either negotiation or legal channels. Consequently, effective and reliable security patrols are essential to upholding the company's vision and mission. To facilitate the selection of the best tower security patrol, a Decision Support System (DSS) is required; such a system evaluates relevant criteria to assist, accelerate, and streamline the decision-making process, ensuring objectivity and accuracy. A DSS is a computer-aided system that gathers data to support the selection of the optimal security patrol. To assist in this selection process, the author combines the Rank Order Centroid (ROC) method with the Multi-Attribute Utility Theory (MAUT) method. The ROC method is chosen because it assigns weights to criteria based on rankings reflecting their respective levels of importance; it offers the advantage of a relatively simple process compared to other weighting methods. Meanwhile, the MAUT method enables rapid calculation—without the need to convert criterion weights into fuzzy numbers—and provides objective recommendations for selecting the best tower security patrol based on company qualifications.

Analisis Segmentasi Pelanggan E-Commerce menggunakan Metode RFM dan K-MEANS Clustering untuk Strategi Bisnis

2026-09-30T13:43:37+07:00

The rapid development of the e-commerce industry has generated vast amounts of customer transaction data. However, most companies face challenges in processing this raw data into strategic business insights. Consequently, management struggles to understand variations in transactional behavior, predict buyer preferences, and maintain personalized customer loyalty. This condition risks increasing customer churn and driving promotional cost inefficiencies due to misdirected marketing efforts. This study aims to perform customer segmentation based on transaction data to provide targeted and measurable marketing strategy recommendations. The methodology integrates the RFM (Recency, Frequency, Monetary) model with the K-Means Clustering algorithm. A secondary dataset sourced from Kaggle was processed through Exploratory Data Analysis (EDA), data cleaning, and data type transformation. The optimal number of clusters was determined using the Elbow Method and validated with a quantitative evaluation metric, the Silhouette Score. The test results indicated that a four-cluster solution provided the best structure with a Silhouette Score of 0.43. The four main identified clusters include Lost Customers (39.2%), Low-Value Customers (37.2%), Big Spender Customers (20.4%), and Loyal Customers (3.2%). These segmentation results were subsequently visualized through an interactive Looker Studio dashboard. The implementation of this dashboard provides practical contributions by delivering operational insights for executing segmented promotional strategies, optimizing marketing budget allocation, and systematically mitigating customer churn risks.

Analisis Sentimen Komentar YouTube Kualifikasi Piala Dunia 2026 Ronde 4 Timnas Indonesia menggunakan Naïve Bayes

2026-10-09T08:58:08+07:00

Public interest in the Indonesian National Team’s 2026 World Cup Qualifiers is evident from the diverse responses expressed in YouTube comments. The diversity of these responses indicates a range of sentiments toward the Indonesian National Team’s performance, making it necessary to cluster the comments to identify emerging sentiment trends. This study aims to analyze the sentiment of YouTube comments regarding the Indonesian National Team during Round 4 of the 2026 World Cup Qualifiers using the Multinomial Naïve Bayes algorithm. A total of 17,404 comments were collected via the YouTube Data API v3 from the Indonesian National Team and RCTI Sports channels. The data then underwent preprocessing, labeling using the InSet Lexicon verified by linguists, weighting using Term Frequency-Inverse Document Frequency (TF-IDF), and classification with the Multinomial Naïve Bayes algorithm. Testing was conducted using three data split scenarios 80:20, 75:25, and 70:30 with the metrics of accuracy, precision, recall, and F1-score. The test results showed that the 70:30 data split yielded the highest accuracy of 79.09%. In that scenario, the negative class achieved a precision of 0.81, a recall of 0.93, and an F1-score of 0.87. The negative class was also the most dominant sentiment, while the number of data points in the neutral and positive classes was lower. These results indicate that Multinomial Naïve Bayes can be used to classify YouTube comments.

Penerapan Convolutional Neural Network Berbasis Computer Vision untuk Identifikasi Varietas Padi Berdasarkan Citra Bulir Gabah

2026-10-13T11:56:23+07:00

Rice variety identification based on paddy grain images faces a challenge because several varieties have relatively similar visual characteristics, particularly in shape, size, and color, which may lead to inconsistent manual identification. This condition indicates the need for a computer vision approach that can automatically learn the visual characteristics of paddy grains. This study aimed to develop a Convolutional Neural Network (CNN) model to identify five rice varieties available at the Agricultural Research and Modernization Center of Gorontalo (BRMP Gorontalo), namely Ciherang, Inpari 30, Inpari 42, Inpari IR Nutri Zinc, and Mekongga. The dataset was obtained through direct image acquisition, consisting of 28 RAW photographs for each variety and nine cropped images from each photograph, resulting in 1,260 images. The research stages included preprocessing, Custom CNN development, evaluation of image size, learning rate, number of convolutional layers, and data splitting, as well as comparison with MobileNetV2, ResNet50, DenseNet121, and InceptionV3. The best configuration used an input size of 224×224 pixels, a learning rate of 0.0005, four convolutional layers, and an 80:20 data split. Testing on 135 final testing images achieved 93.33% accuracy, 93.44% precision, 93.33% recall, and 93.32% F1-score. These results indicate that the Custom CNN effectively identified the five rice varieties under the dataset and experimental conditions used in this study.