Yuen Chark SeeJin En TehZhi Yin GooJun Kit FoongZhen Hui Thean2025-09-302025-09-302025-02-0710.1109/CSPA64953.2025.10932990https://dspace-cris.utar.edu.my/handle/123456789/11415Accurate indoor positioning is essential for location-based services in environments where Global Positioning System (GPS) signals are unavailable. This study investigates an Indoor Positioning System (IPS) based on Visible Light Communication (VLC), enhanced with machine learning techniques to improve positioning accuracy. The performance of two Light Emitting Diode (LED) configurations: 4-LED and 8-LED, was evaluated across spatial layouts of 16 m2, 25 m2, and 36 m2, Using Received Signal Strength (RSS) data as input, Random Forest (RF), K-Nearest Neighbors (KNN), and Extreme Gradient Boosting (XGBoost) models were tested for positioning accuracy. The results show that the XGBoost model outperformed others, achieving the lowest Root Mean Squared Error (RMSE) of 0.0199 m, Mean Absolute Error (MAE) of 0.0150 m, and Mean Positioning Error (MPE) of 0.0237 m in a 16 m2 area with an 8-LED setup. Furthermore, transitioning from a 4-LED to an 8-LED configuration improved RMSE by 48.71 % in 16 m2 area, highlighting the critical role of increased LED density. These findings demonstrate the potential of integrating machine learning with optimized LED configurations to enhance IPS accuracy in diverse indoor environments. © 2025 IEEE.enIndoor Positioning System (IPS)Light Emitting Diode (LED)Machine Learning (ML)Visible Light Communication (VLC)FluorescenceMean square errorPhosphorescenceIndoor positioningIndoor positioning systemLight emitting diodeLightemitting diodeMachine learningMachine-learningPositioning accuracyPositioning systemVisible lightVisible light communicationLight emitting diodestext::conference output::conference proceedings::conference paper