Inform : Jurnal Ilmiah Bidang Teknologi Informasi dan Komunikasi

Integration of Google Maps with the YOLO Method for Urban Greening Monitoring

Yennimar Yennimar — 2026-01-05

The development of artificial intelligence (AI) and computer vision technology has opened up new opportunities in environmental monitoring, especially green open spaces (RTH) in urban areas. This study aims to develop an automatic tree-detection system for satellite imagery using the You Only Look Once version 8 (YOLOv8) algorithm integrated with the Google Maps API. Therefore, the novelty of this study lies in integrating the YOLOv8 model with the Google Maps JavaScript API to produce an interactive spatial visualization of detection results, enabling real-time analysis of vegetation distribution to support sustainable urban policies. The methods used include collecting a 640 × 640-pixel-resolution satellite image dataset from 21 sub-districts in Medan City, object labelling using the Roboflow platform, data augmentation to increase diversity, and model training on Google Colaboratory. The resulting YOLOv8 model achieved excellent performance, with a precision of 0.864, recall of 0.788, and an [email protected] of 0.938, indicating strong object detection capabilities. The detection results were then converted to geographic coordinates and visualized using the Google Maps JavaScript API to provide interactive spatial information on tree distribution in urban areas. This study not only demonstrates the technical feasibility of integrating the YOLOv8 algorithm with the Google Maps API but also provides a methodological framework for spatial AI coupling that can be replicated across other environmental monitoring domains. In addition, the study identifies several limitations, including dependence on image resolution, limited temporal coverage, and relatively lower [email protected]–0.95 performance in dense vegetation, which will guide future model improvement.

Enhancing Intrusion Detection Systems by Integrating Support Vector Machine and Random Forest Classifiers Using Dempster-Shafer Theory

Bilal Waheed — 2026-01-09

The role of Intrusion Detection System (IDS) in defending the networks against the increasing stages of cyberattacks is critical. However, single machine learning models tend to be ineffective due to data imbalances, variability in attack patterns, and ambiguity in network traffic. To address these issues, this paper introduces a hybrid IDS architecture based on Support Vector Machine (SVM) and Random Forest (RF) classifiers, with evidence fusion using the Dempster-Shafer Theory (DST). The strategy leverages the high-quality boundary detection of SVM and the strength of an ensemble of RF. At the same time, DST offers a principled approach to dealing with uncertainty and integrating conflicting evidence. As a benchmark, the CSE-CIC-IDS2018 dataset, a set of labelled benign and multiple-attack traffic flows, was used. Stratified train-test partitioning with a fixed random seed, feature standardization, and class balancing using the Synthetic Minority Oversampling Technique (SMOTE) were considered data pre-processing steps for RF. An SVM trained and optimized with RAPIDS cuML on GPUs was trained and optimized via a grid search over selected hyperparameters. To unify the strengths of models, we further integrated classifier outputs using Dempster-Shafer Theory (DST), which transforms probabilistic outputs into belief assignments and yields an ultimate decision based on the belief assignment with the highest value. The models exhibit high predictive ability, as demonstrated by the experimental results. The DST-based fusion outperformed the two individual classifiers, achieving 97.84% accuracy, 97.41% precision, 94.96% recall, and 96.17% F1-score. In this paper, we show that combining classifiers using DST results in a substantial, computable gain over single-model methods. It is the novelty of using DST-based IDS fusion as well to enhance robustness and balanced detection. These results confirm the value of DST-based fusion in improving IDS performance. These results confirm that combining SVM and RF with DST yields a more robust, reliable IDS. In addition to improving the capability for precise threat discovery, this approach also has certain implications for uncertainly evolving network circumstances, highlighting its suitability for actual-world cybersecurity applications.

The Development of a Type-2 Fuzzy Algorithm for Quadplane Attitude Control System in VTOL to Cruising Transition

Mei Laillatul — 2026-01-09

The Quadplane uncrewed aerial vehicle (UAV) is a combination of a quadcopter system and a conventional aircraft. The Quadplane UAV has three phases: vertical take-off, transition, and cruise. In the transition phase, the Quadplane Tilt-rotor tilts the two front motor axes for forward propulsion. During this transition phase, the aircraft’s balance changes, potentially causing it to crash. This study proposes using a type-2 fuzzy control method. The type-2 fuzzy control method is better at handling uncertainties in the Quadplane, known as the Footprint of Uncertainties (FOU), than the type-1 fuzzy method. In this study, simulations were conducted using MATLAB Simulink, and the results of the type-2 fuzzy control method and the PID method from previous studies were compared. The results of the z-position tracking response using the type-2 fuzzy method yield a rise time of ±3 s, an overshoot of <2%, and a steady-state error of ±0.5 m. The results of the x-position tracking response using the type-2 fuzzy method yield a rise time of ±2.5 s, an overshoot of almost 0%, and a steady-state error of <0.2 m. The results of the Quadplane pitch angle position tracking response using the type-2 fuzzy method produce a rise-time value of ±1.5 s, overshoot ±0.05°, steady state error ±0.02Overall, the type-2 fuzzy controller is proven to be more effective, accurate, and efficient in controlling the hybrid Quadplane in the transition phase, so it is worthy of being implemented in a real prototype with hardware-in-the-loop testing as further research.

Hybrid Face Recognition System Using Haar Cascade and Local Binary Pattern Histogram for Automatic Smart Door Access

Zahra Humaira Kudadiri — 2026-01-29

Conventional key-based door access systems remain vulnerable to loss, duplication, and misuse, motivating biometric access mechanisms that can operate reliably under constrained resources. This paper proposes a hybrid face recognition system for automatic smart door access that integrates Haar Cascade for real-time face detection and Local Binary Pattern Histogram (LBPH) for lightweight recognition. The system adopts an IoT-assisted architecture in which an ESP32-CAM captures facial frames and transmits them via Wi-Fi to a processing unit that performs detection, recognition, and access-decision logic, subsequently activating a solenoid lock and generating Telegram-based event logs to support traceability. Haar Cascade efficiently localizes facial regions via multi-scale scanning and cascade-based rejection, while LBPH encodes local texture patterns into histograms for identity matching with low computational overhead. The proposed system was evaluated with three enrolled participants across seven trials, covering illumination variation, multiple faces, and a controlled failure case designed to simulate an unauthorized or non-verifiable attempt. Experimental results show a successful unlocking rate of 85.71% (6/7 trials) with an average end-to-end response time of approximately 2.20 s, demonstrating that the system can operate within practical latency constraints for real-time access control. In addition, the controlled failure case was correctly rejected, indicating a conservative security posture that prevents unsafe unlocking when facial evidence is insufficient. Overall, the findings suggest that the proposed classical-method hybrid design provides an effective balance between recognition reliability, latency, and deployment feasibility for IoT-enabled smart door security applications, particularly in cost- and power-constrained environments.

Integrating Built-in Feature Importance and Mutual Information for Efficient Android Malware Classification Model

Lukmanul Hakim — 2026-01-29

The rapid growth of the Android operating system in the global market demands efficient and accurate malware detection solutions. This study proposes an Android malware classification approach based on machine learning with a focus on feature selection optimization to achieve an optimal balance between performance and computational efficiency. Using the CICMalDroid2020 dataset, consisting of 11598 samples and 470 dynamic features, this study evaluates four machine learning algorithms (LightGBM, XGBoost, Random Forest, and K-Nearest Neighbours) combined with two feature selection methods: Mutual Information (MI) and Embedded Feature Importance (FI). Experiments were conducted with automatic feature selection over 50-475 features to identify the optimal configuration for each model. The research results show that LightGBM with Feature Importance achieves the best performance, with an accuracy of 96.49%, an F1-score of 95.74%, using only 270 features (a 42.6% reduction), and the fastest test time of 0.036 seconds. XGBoost FI achieves 96.31% accuracy with 225 features (52.1% reduction), Random Forest MI achieves 95.62% with 240 features, while KNN MI achieves 91.37% with 135 features. Feature overlap analysis reveals that the 135 core features selected by KNN MI are a subset of features from other models, with dominant categories including system calls (40%), Android API (25%), network operations (15%), file system patterns (10%), and behavioural patterns (10%). This research shows that the Feature Importance method from tree-based algorithms outperforms Mutual Information by 5-6% in capturing non-linear dependencies and complex interactions in malware behaviour. Feature Importance can detect contextual patterns, such as the combination of getDeviceId and NETWORK_ACCESS, that are only dangerous when occurring simultaneously, which are more easily detected by tree-based methods. The optimal range of 225-270 features provides a sweet spot between comprehensiveness and efficiency; XGBoost with 225 features is only 0.18% below LightGBM but 16.7% more computationally efficient, making it ideal for real-time scanning. The main contribution of this research is the development of a light-weight yet reliable model without destructive sampling techniques, providing a practical solution for real-time malware detection on Android devices with limited resources. This approach successfully reduces dimensions by up to 52% while maintaining, or even improving, performance, making significant contributions to the development of efficient, accurate, and applicable Android malware detection techniques for real-time security systems. For further development, exploring LightGBM-XGBoost ensembles could increase accuracy beyond 97%, along with advanced feature engineering and periodic evaluation of the latest malware variants.

Design and Evaluation of an Integrated BI Solution with Centralized Data Architecture

Anastasia Lidya Maukar — 2026-01-29

This study develops an integrated Business Intelligence (BI) dashboard to address inefficiencies in monitoring key performance indicators (KPIs) at a toy manufacturing company. The previous system required monitoring 48 separate dashboards, used over 25 data sources, and suffered from inconsistent formats that caused frequent errors and delayed reporting. Using a visualization methodology, KPI data for OEE, Quality Performance, and Scrap were standardized and integrated into a centralized SQL Server database via an automated ETL pipeline. The resulting Power BI dashboard improved operational performance by reducing monitoring time from 107.1 minutes by three workers to 78.7 minutes by one staff member (26% reduction), decreasing data inconsistencies by eliminating redundant fields, and enabling near real-time monitoring. These improvements strengthened decision-making accuracy and provided a scalable blueprint for continuous improvement.

A Modified Hybrid Content-Aware Course Recommendation Model for Moodle-Based Learning Management System

Diva Alfiah Hakim — 2026-01-29

Existing course recommendation systems in Learning Management Systems (LMS) often restrict suggestions to the same categories as a user's previously taken courses, limiting diverse course discovery. To address this, this research developed a personalized, cross-category course recommendation system for a data-constrained institutional LMS. This research adapted the Hybrid Content-Aware Course Recommendation (HCACR) framework, integrating a metadata-based user-interest model, a K-Modes-based demographic characteristic model, and a sequential course history model to mitigate data sparsity and cold-start problems. The system was deployed in a Moodle-based environment and evaluated by 171 users. Experimental results show that the model achieved a precision of 26.78% and a recall of 31.07%, which are reasonable given the data constraints in internal government education contexts. Crucially, the system obtained an excellent System Usability Scale (SUS) score of 86.25, indicating high user satisfaction despite the moderate algorithmic precision. While the reliance on sparse metadata limits semantic richness compared to full-content models, this study demonstrates that a hybrid approach is a feasible and effective solution for enhancing course discovery in institutional settings with limited data access.

X-Means Clustering for UX Evaluation of the Candy CBT Application Using the SUS Instrument

Laila Nurzanah — 2026-01-29

The Candy CBT application at SMKS Pancabudi Medan has been operating for three years without adequate usability reviews, posing a high risk of serious problems in user experience (UX), especially in terms of system responsiveness. Therefore, this study aims to analyze the level of system usability and group users based on three aspects (GUI, Navigation, and Responsiveness) using X-Means Clustering to develop recommendations for improvement. The method used is a data mining-based evaluative study involving 400 respondents of a modified System Usability Scale (SUS) questionnaire, with data processed through reverse scoring and Z-Score Normalization before clustering. The results show that the GUI (4.53) and Navigation (4.51) aspects are rated very good. Still, responsiveness is very low (1.48), becoming a major weakness consistent across all clusters. The X-Means Clustering model automatically determined two optimal clusters, with the most dominant cluster specifically showing extreme dissatisfaction with responsiveness (score 1.08). This study contributes a more granular usability evaluation approach by integrating aspect-based SUS features with X-Means Clustering, enabling more precise identification of critical UX weaknesses and supporting data-driven prioritization of usability improvements in CBT systems. Therefore, the application's usability needs to be significantly improved, and recommendations should focus on enhancing non-functional requirements, particularly system performance and stability, to ensure better multi-device responsiveness, with the cluster showing the highest dissatisfaction as the top priority.

Integrating RAD and Design Thinking for Developing a Web-Based POS and Inventory Management System for MSMEs: A Case Study

Imelda Dimentieva — 2026-01-29

Manual transaction and inventory recording in micro, small, and medium enterprises (MSMEs) often leads to recording errors and delayed decisions. This study developed a web-based point-of-sale (POS) and inventory management system for Fresh Market Klatak to streamline transactions and stock control. Development followed Rapid Application Development (RAD) and was integrated with Design Thinking to elicit user needs and iterate on prototypes rapidly. The system was evaluated using scenario-based functional testing and user acceptance testing (UAT). All functional test scenarios passed (100%). UAT with 16 users produced an overall acceptance score of 93% (Very Good), indicating the system is usable and meets operational requirements. Future work will develop a mobile application and integrate payment-gateway services to improve accessibility and transaction efficiency.

Health, Safety, Environment, and Ergonomics Analysis of Solar Power Systems Using an Adaptive Neuro-Fuzzy Inference System

Dinda Aulia Ilma Shafira — 2026-01-30

Solar energy is recognized as a clean energy source; however, its implementation presents various challenges related to Health, Safety, Environment, and Ergonomics (HSEE) aspects that must be addressed. This study aimed to identify subvariables within the HSEE aspects and to analyze HSEE assessments using the Adaptive Neuro-Fuzzy Inference System (ANFIS) method. The identification results indicate that the health aspect comprises two sub-variables: heat stress and toxic materials. The safety aspect includes three sub-variables: electrical risk, fire hazard, and fall risk. The environmental aspect comprises three sub-variables: ecosystem damage, land use, and material recycling. The ergonomics aspect includes three sub-variables: musculoskeletal injury risk, work posture, and manual handling risk. The ANFIS model was developed from questionnaire data categorized into three risk assessment levels: good, fair, and poor. Model performance was evaluated using the Root Mean Square Error (RMSE) as an indicator of accuracy during both the training and testing phases. The evaluation results show RMSE values for the health variable of 0.0120 (training) and 0.0512 (testing); safety of 0.0232 (training) and 0.1515 (testing); environment of 0.0158 (training) and 0.0548 (testing); and ergonomics of 0.0294 (training) and 0.0327 (testing). The overall RMSE values for the health, safety, environment, and ergonomics models were 0.034, 0.140, 0.045, and 0.025, respectively. This study demonstrates that the ANFIS method can serve as a decision-support tool for systematically and adaptively assessing HSEE performance, thereby improving the health, safety, environmental, and ergonomic aspects of solar power plants.

Systematic XGBoost Pipeline for Phishing Website Detection: Hyperparameter Tuning Approach with Nested Cross-Validation

Najma Prameswari — 2026-02-02

Phishing attacks have become increasingly sophisticated and pose a critical threat to cybersecurity, with more than 4.7 million attacks reported in 2023. Traditional blacklist and rule-based detection struggles to keep pace with evolving URL patterns and impersonation techniques. Rather than proposing a new classifier, this study presents a systematic and reproducible XGBoost-based phishing detection pipeline intended as an academic baseline with operationally motivated evaluation (not a production-integrated system). The Mendeley Phishing Websites dataset (58,645 URLs; 30,647 phishing and 27,998 legitimate) with 111 URL- and website-based features. The pipeline applies data cleaning, column-transformer-based pre-processing, and a stratified 80:20 train–test split, with all pre-processing steps fit on the training data only to reduce leakage risk. The final model uses 98 active features after removing 13 constant attributes; quasi-constant features are analyzed and retained. Continuous features are sanitised, log-transformed, and standardised, while binary features are left unchanged. Hyperparameters are tuned via stratified cross-validation using the ROC-AUC metrics, followed by early stopping, probability calibration, and simple threshold tuning. On the hold-out test set, the optimized model, set at a 0.50 decision threshold, achieves 96.34% accuracy, 96.31% precision, 96.70% recall, and 96.51% F1-score, improving over a default XGBoost baseline and yielding fewer false positives and false negatives. These results show that a systematically designed XGBoost pipeline provides a strong and reproducible baseline for URL-based phishing website detection and offers a practical foundation for future work on cost-sensitive learning and temporal validation. This study is limited to tabular URL/website feature-based detection and does not include visual content analysis, HTML/DOM parsing, or deep learning on raw text/images.

Performance Analysis of Support Vector Machine with Hyperparameter Tuning for Engagement Rate Classification of TikTok Digital Marketing Content

Muhammad Dzar Algifahri — 2026-02-02

The effectiveness of TikTok digital marketing content often varies, creating challenges for marketers in achieving consistent audience engagement. Differences in posting time, upload frequency, video length, and music selection make it difficult to distinguish engagement rate (ER) levels accurately. This paper investigates key factors affecting engagement and proposes a classification framework to separate high- and low-engagement TikTok content. A machine learning approach using the Support Vector Machine (SVM) algorithm is applied to a dataset of TikTok videos collected between 2023 and 2025. From an initial dataset of 2,992 videos, 1,638 representative samples were retained after data cleaning and creator-level filtering. The research process involves feature engineering, engagement-based labelling with a 5% ER threshold, data normalisation, and dataset partitioning with an 80:20 training–testing split. The baseline SVM model achieved an accuracy of 70.43%, indicating limited ability to distinguish low-engagement content. After systematic hyperparameter tuning, the optimised linear SVM model demonstrated improved performance, achieving an accuracy of 88.41% with an optimal regularisation parameter (C = 100) and more balanced classification results. Model interpretation indicates that video duration, temporal attributes, and audio characteristics play important roles in separating engagement levels. The proposed framework is intended for post-hoc engagement classification rather than engagement prediction, providing interpretable insights to support TikTok digital marketing strategy optimization.

Sentiment Analysis and Emotional Reviews of Hospital Services Using Naïve Bayes and Support Vector Machine (SVM)

Muhammad Rio Lintang Cahya — 2026-02-03

Hospitals are crucial healthcare institutions that play a vital role in providing medical services to the community. Patient perceptions and experiences regarding hospital services are often reflected in reviews left on digital platforms such as Google Maps. This study aims to analyze public sentiment and emotions toward hospital services in Semarang City using Google Maps user reviews. A total of 16,364 reviews from 21 hospitals were collected between 2023 and 2024. Sentiment labelling was performed using a lexicon-based approach to classify reviews into positive, negative, and neutral categories. To explore the emotions expressed in the reviews, the NRC Emotion Lexicon (EmoLex) was used to identify eight basic emotions. The analysis revealed that 'Trust' was the most dominant emotion (7,090 words), indicating high patient confidence, followed by 'Joy'. Furthermore, for predictive modelling using Naïve Bayes and Support Vector Machine (SVM) algorithms, SVM achieved 90% accuracy, whereas Naïve Bayes achieved only 78%. The results of this analysis are expected to serve as input for hospitals to improve service quality and as a reference for prospective patients in selecting a hospital.