Jurnal Sains, Nalar, dan Aplikasi Teknologi Informasi

Web-Based Case File Management System for Motor Vehicle Theft Crimes

Cut Fiarni — 2026-01-16

This study addresses the inefficiencies in managing motor vehicle theft case data at the West Java Police Department by developing a web-based file management system. The current manual process, which relies on Microsoft Excel, is prone to human error, data duplication, and reporting delays. Our research introduces a system that automates data categorization, detects data similarities (such as duplicate chassis or engine numbers), and streamlines the verification and validation process, which was previously a time-consuming manual task. By employing object-oriented programming principles, the system accommodates diverse data types and dynamic reporting needs. The system's novelty lies in its specific focus on vehicle theft cases and the integration of a multi-level verification process. User Acceptance Testing (UAT) using the UTAUT model showed high user acceptance, with a behavioral intention of 87.5%, performance expectancy of 84%, and effort expectancy of 82%. This new system significantly improves data accuracy, accelerates reporting, and enhances the overall efficiency of criminal case handling.

Performance Comparison of Adam and SGD Optimizers in Transfer Learning Based CNN for Banana Leaf Disease Classification

Zaid Romegar Mair — 2026-01-16

Banana leaf diseases significantly reduce crop productivity, yet automated detection systems based on deep learning often rely on limited datasets, where training stability and generalization become critical challenges. Although Convolutional Neural Networks (CNNs) have been widely applied for plant disease classification, systematic comparisons of optimization algorithms under small dataset conditions remain limited, particularly for banana leaf disease identification. This study addresses this gap by comparing the performance of Adaptive Moment Estimation (Adam) and Stochastic Gradient Descent (SGD) optimizers within a transfer learning–based CNN framework. Six pre-trained architectures VGG16, VGG19, ResNet50, DenseNet121, MobileNet, and NASNetMobile were evaluated using 1,652 annotated banana leaf images classified into Sigatoka, Cordana, Pestalotiopsis, and healthy leaves. Both optimizers were trained under identical experimental settings to ensure a fair comparison. Experimental results show that VGG19 achieved the highest accuracy, reaching 85% with Adam and 83% with SGD, while lightweight architecture exhibited lower performance due to underfitting. The findings demonstrate that optimizer selection plays a crucial role in improving CNN performance for banana leaf disease classification, especially when data availability is limited.

Strategic Information Systems Planning Using the Tozer Methodology: A Comprehensive Literature Review

Fajar Indra Nur Alam — 2026-01-16

The rapid advancement of Information Systems (IS) and Information Technology (IT) has prompted organizations across sectors to adopt systematic approaches for aligning technological initiatives with business strategies. One of the most widely applied frameworks in Indonesia and beyond is the Tozer Methodology for strategic information systems planning. This paper presents a comprehensive literature review of ten selected studies applying the Tozer framework in diverse organizational contexts, including education, telecommunications, publishing, microfinance, trade, media, and interior design services. Through critical synthesis, this review identifies common analytical tools used alongside Tozer (SWOT, PEST, Value Chain, CSF, McFarlan’s Grid, Five Forces), evaluates the effectiveness of the methodology, and highlights recurring challenges such as integration issues, data duplication, and limited scalability. While findings consistently affirm Tozer’s practicality and adaptability, gaps remain in integrating Tozer with emerging technologies such as cloud computing, big data analytics, and artificial intelligence. This review contributes by mapping trends, identifying research gaps, and providing recommendations for future studies and organizational practices aiming to optimize IS/IT strategic alignment.

An Explainable Spatio-Temporal Decision Support System (DSS) Using XGBoost And SHAP For Urban Complaint Trend Prediction

Moeng Sakmar — 2026-01-16

The increase in the volume of public complaints in urban areas requires an accurate and explainable decision support system. This study developed an Explainable Decision Support System (xDSS) based on the Extreme Gradient Boosting (XGBoost) algorithm combined with the SHapley Additive Explanations (SHAP) method to predict spatial and temporal trends in public complaints in DKI Jakarta Province. The research data was obtained from the Satu-Data Jakarta portal and included multi-year complaint reports that were processed through aggregation, temporal feature engineering, and regression-based metric evaluation. The results show that the XGBoost model has high predictive performance with an R² value of 0.8425, MAE of 2.9858, and RMSE of 4.9928, indicating the model’s ability to explain more than 84% of the variation in the actual number of complaints. SHAP analysis revealed that temporal features such as complaint_lag1 and complaint_ma3 had the most dominant influence, while external variables such as rainfall (rainfall_mm) and population density (population_density) also made positive contributions. These results indicate that the dynamics of public complaints are influenced by a combination of historical factors and environmental conditions. Practically, this xDSS system can provide accurate predictions and transparent interpretations, thereby supporting the implementation of Smart Governance and evidence-based policy. This approach strengthens the application of Explainable Artificial Intelligence (XAI) in public service governance by providing accurate, ethical, and auditable models to support strategic decision-making in the era of digital government.

Comparative Evaluation of Federated Learning Algorithms in Dirichlet Non-IID Medical Imaging

Michael Angello Qadosy Riyadi — 2026-01-16

Machine learning has achieved diagnostic performance comparable to clinical experts on medical imaging, yet centralized training paradigms necessitate patient data aggregation, risking violations of privacy regulations such as GDPR and HIPAA. In 2023, 1,853 healthcare data breaches were reported in the United States, compromising over 133 million medical records, rendering raw inter-institutional data exchange increasingly unsustainable. Federated Learning (FL) offers a viable solution by enabling collaborative model training without data transfer. However, prior studies predominantly evaluate single algorithms and often neglect non-IID Dirichlet-distributed conditions and probabilistic calibration metrics like log-loss. This study rigorously compares FedAvg, FedProx, FedSVRG, and FedAtt across three MedMNIST v2 datasets—PneumoniaMNIST (binary), DermaMNIST, and BloodMNIST (multi-class)—using three clients under non-IID Dirichlet partitioning (α=0.1) over 50 communication rounds. FedProx demonstrates the most consistent performance and stability, achieving accuracy of 0.9521 and log-loss of 0.1850 on PneumoniaMNIST; 0.8595 and 0.4066 on BloodMNIST; and 0.5747 and 1.5996 on DermaMNIST. It also exhibits fastest convergence and superior probability calibration. Thus, FedProx’s proximal regularization enhances FL robustness against extreme clinical heterogeneity, establishing it as a scalable, privacy-preserving framework for cross-institutional medical image diagnostics.

Real-time Forensic Reconstruction of IPv6 NA Flood Attacks: A D4I Approach

Frendi Yusroni Romadhona — 2026-01-17

The global transition to IPv6 has introduced new attack surfaces within core network protocols, particularly the Neighbor Discovery Protocol (NDP). One of the most critical yet often overlooked threats is the Neighbor Advertisement (NA) Flood attack. Unlike conventional volumetric DDoS attacks aimed at saturating network bandwidth, NA Flood exploits the Stateless Address Autoconfiguration (SLAAC) mechanism to trigger resource exhaustion on target devices. Investigating such incidents presents unique forensic challenges, as attack traces in volatile memory are often lost when using traditional dead forensics methods. This study implements a real-time forensic investigation approach by integrating Live Forensics methods with the Digital Forensic Framework for Reviewing and Investigating Cyber Attack (D4I). This method is applied to acquire crucial volatile artifacts during the attack and reconstruct the modus operandi through Cyber Kill Chain (CKC) mapping and Chain of Artifacts (CoA) construction. Experimental results demonstrate that NA Flood attacks possess dangerous asymmetric characteristics: generating low network traffic (4.71 Mbps) while causing a CPU surge of up to 50% and a memory increase of 89.5 MB on the target server. The novelty of this study lies in the integration of Live Forensics with the D4I framework to acquire volatile data in real-time and systematically transform raw artifacts into a comprehensive forensic conclusion. This approach successfully reconstructs the 5W1H (Who, What, Where, When, Why, How) elements of the incident and visualizes the shift of the point of failure from the network infrastructure to the endpoint, offering a robust model for investigating protocol-based resource exhaustion attacks.

A Systematic Review of Convolutional Neural Network Models for Tomato Leaf Disease Detection

Fiki Sanora — 2026-01-17

Tomato leaf disease can cause a decline in productivity and crop failure, making early detection very important in precision farming practices. Manual detection methods, which are still commonly used in the field, have limitations in terms of speed and accuracy, requiring an automated image-based approach. Convolutional Neural Networks (CNNs) have become a leading technique in plant disease classification, but the diversity of architecture used requires systematic study to identify the most effective model. This study summarizes, compares, and evaluates CNN models for tomato leaf disease detection through a Systematic Literature Review (SLR) that adopts the PRISMA guidelines, covering the stages of identification, screening, feasibility assessment, and inclusion. A search in Scopus (2022–2025) using the query: (“Convolutional Neural Network” OR ‘CNN’) AND (‘tomato’ AND “leaf disease detection”) yielded 21 relevant articles. Analysis shows common preprocessing such as image resizing, data augmentation, and denoising. The best CNN architecture is InceptionV3 (most frequently used and high performing), followed by DenseNet201, MobileNetV2, and ResNet152V2. Architectures with optimal depth and high computational efficiency are preferred. This study provides a comprehensive map of CNN models to support architecture selection in tomato leaf disease detection. Future research directions include improving image quality, integrating attention mechanisms, semantic segmentation, and developing concise and efficient models for field applications.

Implementation Layered Mitigation Techniques for Unrestricted File Upload and Server-Side JavaScript Injection

Salman Akbar Hasbullah — 2026-01-18

The popularity of Node.js as a server-side application development platform has introduced new security challenges stemming from the dynamic features of JavaScript. Vulnerabilities such as Unrestricted File Upload (UFU) and Server-Side JavaScript Injection (SSJI) often arise from insecure input handling and over-reliance on third-party libraries. This research aims to design, implement, and evaluate a multi-layered security mitigation model for Node.js-based web applications built using the Express.js framework. A constructive research approach was employed, wherein hybrid security middleware was developed to enforce comprehensive validation. This middleware integrates content-based file type validation (magic numbers), file name sanitization to prevent path traversal, and malicious input pattern blocking to mitigate SSJI and prototype pollution. The effectiveness of the model was empirically evaluated within a controlled local testing environment using the Jest testing framework by comparing a vulnerable application against its secured counterpart. Test results demonstrate that the proposed mitigation model successfully blocked 100% of the tested attack scenarios, achieving 100% test code coverage on the core security logic. This research yields a practical solution capable of enhancing the resilience of Node.js applications against common attacks exploiting language-specific features