High-Dimensional Statistical Feature-Enhanced BLE Fingerprinting Indoor Localization

Indoor localization based on Bluetooth Low Energy (BLE) fingerprinting often suffers from accuracy degradation due to fluctuations in the Received Signal Strength Indicator (RSSI) and environmental dependencies. To mitigate these effects, this study proposes an advanced indoor localization framework that leverages high-dimensional statistical features of RSSI. Departing from traditional methods centered on mean RSSI values, our approach integrates five distinct metrics—mean, variance, maximum, minimum, and interquartile range (IQR)—to capture the complex temporal fluctuations and spatial signatures of RSSI at each reference point. Furthermore, a Back-Propagation Neural Network (BPNN) correction model is developed to learn the relationship between these observed statistical features and the ideal RSSI values derived from the Log-Normal Shadowing Model (LNSM). Experimental results obtained in a real-world indoor environment demonstrate that the proposed method significantly reduces localization errors, particularly in challenging areas such as near walls and corners. The evaluation confirms that the proposed approach improves localization accuracy across multiple algorithms, including k-Nearest Neighbor (k-NN), Weighted k-NN (WKNN), and Support Vector Regression (SVR). Notably, the proposed method achieves a mean localization error of 1.391 m and reduces the 95th percentile error by approximately 41.5% compared with conventional methods without correction.