JITSI : Jurnal Ilmiah Teknologi Sistem Informasi

Transformer-Based Sentiment Analysis of Mpox Discourse on Twitter for Digital Epidemiology

2026-10-14T11:49:54+01:00

The Mpox outbreak has sparked extensive public discourse on social media platforms like Twitter (X), where users express emotions ranging from worry and suspicion to support. Understanding these emotions can be of great help to public health authorities attempting to communicate and respond while preparing for outbreaks like this one. However, conventional sentiment analysis approaches often fall short, as prior algorithms have failed to recognise syntactic categories in informal, almost carefree social media text. The study was carried out to evaluate the efficacy of transferrable transformer-based technologies in sentiment analysis of Twitter texts concerning the Mpox crisis. The researchers gathered 6,377 tweets using Apify scraping software from the Web and further filtered them to 5,142 English-language tweets after removing duplicates and non-English tweets. Sentences were classified using the VADER tool, then used to train a transformer-based classification model with AutoGluon's MultiModalPredictor, employing the pre-trained ELECTRA architecture. The model was then trained and tested with an 80:20 training-testing split and achieved an overall accuracy of 83.92%, with macro-averages of 79.78%, 79.67%, and 79.57% for precision, recall, and F1-Score, respectively. The comparison between the transformer and the VADER baseline was based on the transformer's significant leap in sentiment classification; it exhibited clear improvements. To demonstrate the practical applicability of the models, the trained model was deployed via an integrable Streamlit dashboard, where users could upload tweet datasets for the classifier to generate sentiment predictions. Some evaluations of sentiment trends were also conducted. The study also sheds light on how businesses based on transformer models have been capitalised to investigate large-scale social discourse and, in turn, to support digital epidemiology and public health monitoring during infectious disease outbreaks.

Cloudflare Tunnel-Based Network Infrastructure for Application Deployment

2026-10-14T11:49:25+01:00

This study designs a secure and efficient network infrastructure using Cloudflare Tunnel to solve the challenge of exposing on-premise K3s services. By applying Zero Trust Network Access (ZTNA) principles and deploying a cloudflared connector inside the cluster, the solution provides stable and secure public endpoints without opening inbound ports. As a result, service management, security, and development/testing workflows become faster

Implementasi Dan Optimasi Lightweight Intrusion Detection System Berbasis Snort Pada Server Berspek Rendah

2026-10-14T14:15:12+01:00

The advancement of information technology has encouraged village governments to implement web-based public service systems to improve service effectiveness. The Sapa Warga Desa Cibungur website was developed as a digital platform that enables citizens to submit aspirations, complaints, and administrative requests online. However, web-based systems are vulnerable to various cyber threats such as SQL Injection, Brute Force attacks, Cross Site Scripting (XSS), and unauthorized access attempts. Therefore, an effective security mechanism capable of detecting suspicious activities in real time is required. This study aims to implement an Intrusion Detection System (IDS) using Snort to improve the security of the Sapa Warga Desa Cibungur website. The software development process follows the Waterfall methodology, consisting of requirement analysis, system design, implementation, testing, and maintenance. The web application was developed using PHP, MySQL, HTML, CSS, and JavaScript, while Snort was

Analisis Efisiensi dan Skalabilitas Network Automation Berbasis Python pada Konfigurasi Multi-Router MikroTik

2026-10-14T11:50:49+01:00

The increasing complexity of computer networks requires a more efficient and consistent configuration process. Manual router configuration using the Command Line Interface (CLI) is time-consuming and prone to human error. This study analyzes the efficiency and scalability of Python-based network automation for configuring multiple MikroTik routers in a GNS3 simulation environment. The automation system utilizes the Netmiko library over Secure Shell (SSH) to perform centralized router configuration. Performance evaluation was conducted by comparing manual and automated configuration methods based on deployment time, configuration consistency, and scalability. The results show that the proposed automation significantly reduces configuration time while maintaining consistent configurations across all routers. In addition, the system demonstrates good scalability by accommodating network expansion with only a slight increase in deployment time. These findings indicate that Python-based network automation is an effective approach to improving efficiency and simplifying multi-router network management.

Improving Support Vector Machine Performance with Hyperparameter Tuning Methods Using a Balanced-Moderate Model for Palm Oil Production Prediction

2026-10-14T11:50:23+01:00

Prediction is an essential step for anticipating future events, encompassing both accuracy and the rate at which predictions are generated. In palm oil production, which holds strategic economic value in the global economy, particularly for Indonesia, precise forecasting is essential. Machine learning algorithms are indispensable for producing accurate predictions, especially when working with large-scale, long-term datasets. This study proposes a novel Balanced-Moderate hyperparameter tuning strategy for Support Vector Regression (SVR) in forecasting global palm oil production. This strategy represents an innovative approach that integrates gradual parameter adjustment with Bayesian optimization to produce moderate and stable predictions. Unlike previous approaches that prioritize extreme parameter search, the Balanced-Moderate method systematically adjusts the values of C (regularization), gamma, epsilon, and kernel through incremental steps, thereby reducing model complexity while enhancing generalization capability. This method was applied to global palm oil production data spanning 1961–2021, comprising 12,053 records from all producing countries. The baseline SVR model, without hyperparameter tuning, yielded an MSE of 0.5969, RMSE of 0.7726, MAE of 0.2207, and R² of 0.3759. After applying the Balanced-Moderate strategy combined with Bayesian optimization, performance improved significantly, achieving an MSE of 0.0006, RMSE of 0.0257, MAE of 0.007, and R² of 0.7937. This result outperformed Grid Search (R² = 0.556) and Random Search (R² = 0.512), confirming the superiority of the proposed approach. These findings hold significant relevance for palm oil supply chain planning and plantation policy formulation, given that forecasting accuracy is a crucial factor in ensuring economic sustainability.

Perbandingan Machine Learning dan Deep Learning untuk Klasifikasi Spam dengan Class Balancing

2026-10-14T11:48:31+01:00

SPAM/HAM classification faces two primary challenges: class imbalance and the need to select a model that is both accurate and efficient. This study compares The Conventional Machine Learning Algorithm Naive Bayes and Support Vector Machine with Recurrent Deep Learning Algorithm Long Short-Term Memory, and Gated Recurrent Unit using a class-balancing strategy. The data underwent cleaning and deduplication before being split into a development set (80%) and a final test set (20%). Balancing was applied exclusively to the training set by downsampling HAM and oversampling SPAM to reach 1,500 samples each, while the validation and final test sets retained their natural distributions. NB and SVM utilized TF-IDF, whereas LSTM and GRU employed tokenization, 128-dimensional embeddings, and padding masks. NB achieved an accuracy of 98.63% and a Macro-F1 score of 96.65%, the highest among the four models. SVM yielded a PR-AUC of 97.55%; GRU achieved the highest SPAM recall at 94.96%; and LSTM attained the highest ROC-AUC at 98.92%. NB's training time was merely 0.004 seconds, compared to 6.61 seconds for GRU and 12.95 seconds for LSTM. The results demonstrate that traditional models remain competitive, offering a superior performance-efficiency trade-off for short-message data following class balancing.

Classification of Avocado Varieties and Maturity Levels Using Quantized MobileNetV2 in a Mobile Application

2026-10-14T11:48:02+01:00

Visual grading of avocado ripeness and variety by distributors and consumers is subjective, and published deep-learning studies on avocado rarely report what happens when a trained model is integrated into a working application. This study presents a proof-of-concept, end-to-end system that classifies two Indonesian avocado varieties (Aligator and Miki) and four ripeness stages (unripe, half-ripe, ripe, overripe) from smartphone photographs. Two MobileNetV2 transfer-learning models were trained on a self-collected image set photographed from several fruits over multiple observation days, with training, validation and test subsets separated by fruit identity to prevent data leakage. The models were converted to post-training-quantized TensorFlow Lite (≈2.7 MB each) and served through a FastAPI–MySQL backend to a Flutter application. On a held-out test set of 32 images, the variety model classified all images correctly (100%; Wilson 95% CI 89.3–100%), and the ripeness model reached 84.4% accuracy (95% CI 68.2–93.1%), a macro-F1 of 0.814, a Cohen's κ of 0.686 and a quadratic-weighted κ of 0.900, against a majority-class baseline of 62.5% (one-sided exact binomial p = 0.006). Four of the five ripeness errors were half-ripe fruits predicted as unripe; half-ripe recall was 0.375 (3/8). Integration testing exposed two defects invisible to offline evaluation, a class-label ordering error and double input normalization, which caused near-constant predictions until corrected. Because the test set is small and drawn from a limited number of fruits, the results should be read as preliminary, and larger fruit-level evaluation and field testing with users are needed.

Perancangan User Interface Aplikasi Pembelajaran Antariksa Virtual Reality Menggunakan Metode Child Centered Design

2026-10-14T11:47:35+01:00

Virtual Reality (VR) has the potential to improve learning quality through interactive and immersive learning experiences, particularly in introducing space science to elementary school students. However, the user interface design of VR-based learning applications often does not fully consider children’s characteristics. This study aims to design a child-friendly Virtual Reality learning application user interface using the Child Centered Design (CCD) method. The design process follows the CCD stages, including context of use analysis, requirement specification, design solution development, and evaluation. Data were collected through observation, literature study, and design evaluation. The results indicate that applying Child Centered Design produces a user interface that is easy to use, visually appealing, and appropriate for children’s needs. In conclusion, the CCD method effectively supports the development of child-friendly VR learning applications for space exploration.

Telkom Indonesia's Daily Reporting Business Process Reengineering Using Throughput Efficiency

2026-10-14T11:47:08+01:00

This study aims to evaluate and redesign operational business processes in the Area Network Division (Arnet) of PT Telkom Indonesia, Malang region, using the Business Process Reengineering (BPR) approach. Four processes were analyzed: daily reporting, equipment maintenance, module troubleshooting, and device installation. Process mapping was conducted using the American Society of Mechanical Engineers (ASME) standard to classify activities, and process performance was measured using throughput efficiency, defined as the ratio of value-added activity time to total cycle time. The results indicate that the daily reporting process had the lowest efficiency at 15.82% with a total cycle time of 158 minutes, while the maintenance process achieved 77.9% efficiency with a total cycle time of 136 minutes. Based on BPR principles, redesign efforts were focused on processes with significant inefficiencies, namely daily reporting and maintenance. The reporting process was redesigned through a Telegram Bot based automation mechanism, reducing total cycle time to 20 minutes and increasing efficiency to 75% (an improvement of 59.18%). Meanwhile, the maintenance process was improved by restructuring activity sequences to eliminate idle time during filter drying, reducing total cycle time to 114 minutes and increasing efficiency to 93.0% (an improvement of 15.1%). The findings demonstrate that BPR implementation supported by ASME process mapping and throughput efficiency measurement effectively identifies non-value added activities and produces measurable and significant process improvements.

Design of a Web-Based Practicum Management System Using the Prototyping Method

2026-10-14T11:46:41+01:00

Practicum management at the Department of Informatics, Universitas Siliwangi is still conducted manually using physical documents, resulting in registration queues, delayed schedule information, risk of losing assistance logs, and inaccurate grade recalculations. This study aims to design a web-based Practicum Management System using the Prototyping method, covering requirements gathering, quick design, building prototype, evaluation, and prototype refinement. The system integrates four main actors (Students, Assistant Lecturers, Supervising Lecturers, and Laboratory Administrators) through ten functional use cases, built upon a Three-Tier architecture that separates the presentation, application logic, and data layers. The novelty lies in automating laboratory business rules within the application logic layer: a server-side timestamp validation algorithm that enforces submission deadlines and calculates a proportional lateness penalty, combined with an automatic final-grade calculation engine operating on a nine-entity database normalized to the third normal form (3NF). This joint integration has not been explicitly addressed in prior web-based laboratory or practicum management system designs. Technical outcomes include a context diagram, a four-process DFD Level 1, and the nine-entity ERD. The resulting high-fidelity prototype was evaluated through Task-Based Usability Testing and a System Usability Scale (SUS) questionnaire involving 25 respondents, yielding an overall average score of 80.25 (82.5 for the Student dashboard and 85.0 for the Laboratory Administrator dashboard), placing the system at an Acceptable acceptance level with an Excellent rating. These results confirm the design is highly intuitive and ready to serve as the primary reference for subsequent system implementation

Evaluasi User Experience pada Aplikasi Platform Sekolah Pintar Mobile Menggunakan UEQ di UNIPDU

2026-10-14T11:46:14+01:00

This study evaluated the user experience of the Platform Sekolah Pintar Mobile application used by students at Universitas Pesantren Tinggi Darul Ulum Jombang. The purpose of this study was to determine the level of user experience and identify the quality of the six dimensions of the User Experience Questionnaire as a basis for application improvement. This study used a quantitative approach with the User Experience Questionnaire instrument consisting of 26 items. Data were collected through questionnaires distributed to 97 respondents from a population of 3097 student users of the application. The collected data were tested using validity and reliability tests, transformed into the User Experience Questionnaire scale from -3 to +3, and analyzed using the User Experience Questionnaire Data Analysis Tool. The validity test showed that all items were valid, while the reliability test produced a Cronbach’s Alpha value of 0.949, indicating that the instrument was highly reliable. The mean results showed that Attractiveness scored 0.799, Perspicuity 0.943, Efficiency 0.844, Dependability 0.682, Stimulation 0.654, and Novelty 0.383. Perspicuity obtained the highest score, while Novelty obtained the lowest score. The benchmark results showed that most dimensions were categorized as Below Average, while Dependability was categorized as Bad. Therefore, the application was relatively easy to understand and fairly efficient, but still required improvement in system reliability, visual appeal, user stimulation, and feature innovation.

Keywords: User Experience; User Experience Questionnaire; PSP Mobile; UI/UX Evaluation; Mobile Application

Penerapan GMM dan ARM untuk Analisis Mobilitas Penumpang TransJakarta

2026-10-14T11:45:44+01:00

TransJakarta passenger mobility shows dynamic patterns influenced by travel time, duration, service-use intensity, and route connectivity. This study constructs probabilistic passenger mobility segments and analyzes inter-corridor travel associations within each segment. The data consist of TransJakarta tap-in and tap-out transactions processed through data quality assessment, time transformation, filtering, feature engineering, missing value imputation, data cleaning, and variable normalization. Passenger segmentation was performed using Gaussian Mixture Model with model selection based on Bayesian Information Criterion, while travel association patterns were analyzed using Association Rule Mining. The final dataset consisted of 168,132 observations, and the selected GMM formed five mobility clusters. The ARM results show that the most numerous and stable association rules appeared in commuter segments, especially Evening Commuter and Early Morning Commuter. These findings indicate that combining GMM and ARM provides a structured understanding of passenger behavior segmentation and inter-corridor travel relationships.