Fuzzy-Enhanced Deep Localization for Robust Indoor Positioning under Structured Noise

Indoor Positioning Systems (IPS) are vital for smart cities and IoT applications but are hindered by Structured Non-Stationary Noise (SNSN) in Received Signal Strength (RSS) data, driven by dynamic indoor factors like human movement, furniture occlusion, and signal reflections in environments such as hospitals and shopping malls.Traditional methods, including k-Nearest Neighbors and manual fuzzy logic, fail to address SNSN's unpredictable patterns, while standalone deep learning models often overfit noisy inputs.We propose Fuzzy-Enhanced Deep Localization (FEDL), a hybrid framework integrating deep learning with fuzzy logic to tackle SNSN.FEDL employs a dual-network architecture: one network automates fuzzy membership derivation from time-series RSS data, while another fuses raw and fuzzified features for precise coordinate prediction.Evaluated on the UJIIndoorLoc dataset, FEDL achieves a mean localization error of 3.79 m, reducing error by 22% compared to baselines.This robust, scalable solution eliminates manual rule design, enabling reliable navigation in dynamic indoor settings.