A Resilient Indoor Localization Framework Using Wireless Signal Sensing and Data-Driven Machine Learning Models

Abstract

Indoor localization has become essential for enabling location-based services in environments where Global Navigation Satellite Systems (GNSS) are unreliable due to signal attenuation and multipath effects. This study presents a robust indoor localization framework that integrates wireless signal sensing with data-driven approaches, including machine learning and deep learning models. Received Signal Strength Indicator (RSSI) measurements are collected from multiple access points to construct a fingerprint-based dataset, which is subsequently pre-processed and used to train Random Forest and Deep Neural Network (DNN) models for position estimation in both two-dimensional (2D) and multi-floor (3D) environments. The proposed system is evaluated under varying conditions, including different access point densities, noise levels, and multi-floor configurations. Experimental results demonstrate that the deep learning model achieves superior performance, with sub-meter localization accuracy (approximately 0.5–0.8 m), outperforming traditional methods such as K-Nearest Neighbours (KNN) and trilateration. Furthermore, the system exhibits strong robustness to RSSI noise and maintains high performance even in sparse deployment scenarios. In multi-floor experiments, the model achieves over 98% floor classification accuracy with minimal vertical error. These findings confirm that data-driven indoor localization provides a scalable, reliable, and accurate solution suitable for smart buildings, industrial IoT, and navigation applications.