Published research at scientific conferences in Computer Science, Machine Learning, and related technology.
Navigation apps such as Google Maps have no direct way to search for Points of Interest (POI) along an entire travel route, only around a single point. The existing RouteBoxer approach runs at O(n³) complexity, making it slow for long routes. This study proposes RouteSegmentation: generating perimeter polygons (a rectangle plus a half-circle) along each route segment, then merging them via polygon union operations, at O(n) complexity.
Building on the RouteSegmentation algorithm, this study adds a POI-detection layer: once the perimeter area along a route is formed, POIs falling inside that area are detected using a point-in-polygon (ray-casting) test, allowing navigation apps to surface relevant places (restaurants, gas stations, etc.) along the whole journey rather than only near the final destination.
Shortest-distance public transit route recommendations (classic Dijkstra) do not reflect real traffic conditions. This study proposes a time-based Dijkstra variant that folds a traffic-speed variable into the graph's edge weights, then compares it against distance-based Dijkstra on a simulated Malang City transit map.
RouteSegmentation had not been tested for intercity travel with far more route points (up to 2,000), where processing cost rises sharply. This study applies Douglas-Peucker line simplification (a 50-meter threshold) before running RouteSegmentation, to cut the point count without sacrificing the route's visual shape.
Hyperparameter tuning for software defect prediction (SDP) models via grid search or random search tends to be inefficient. This study applies Bayesian Optimization to SVM, Random Forest, Gradient Boosting, and Ensemble models, then measures its impact on accuracy and computation time across four public NASA PROMISE datasets.
This study proposes an RGC ensemble model (combining Random Forest, Gradient Boosting, and CNN) for software defect prediction, compared against single models such as RF, GB, SVM, CNN, LSTM, XGBoost, AdaBoost, and KNN across five public NASA PROMISE datasets.
Large medical image datasets (ISIC 2016 skin lesions) slow down machine learning and deep learning model training. This study compares the impact of Huffman Encoding versus Discrete Cosine Transform (DCT) compression on preprocessing time, classification time, and accuracy across CNN, SVM, Random Forest, Logistic Regression, Gradient Boosting, and Ensemble models.
This study compares the performance of four lossy image formats — HEIC, JPG, WebP, and AVIF — when combined with Huffman Encoding across different resolutions, evaluating image quality (PSNR, SSIM, MSE), processing time, and compression ratio to determine the most suitable format for different use cases.
This study examines the influence of gamification and three levels of cognitive abstraction — hedonic value, social value, and utilitarian value — on repurchase intention in the DANA e-wallet application, and tests whether gamification moderates these three values, among 205 undergraduate and postgraduate students at a public university in Malang.
Using the Uses and Gratification Theory (UGT) framework, this study examines how utilitarian gratification (ease of use, self-presentation) and hedonic gratification (enjoyment, passing time) from gamification features affect user satisfaction and continued use intention of the DANA e-wallet app, among 131 undergraduate and postgraduate students at a public university in Malang.