RBF-KAN: Integrated Approach for Accurate Indoor Localization in Dense Grid RSSI Fingerprint

The limitations of satellite signal propagation to penetrate buildings gave rise to indoor localization technology. Indoor localization can enhance navigation services in hospitals, universities, malls, and warehouses. The development of this technology has explored varied approaches, including signal parameters, localization techniques, and machine learning algorithms. Most researchers use the Received Signal Strength Indicator (RSSI) as a signal attribute for localization due to its ease of collection and processing. However, RSSI is susceptible to multipath effects, causing data anomalies. We propose a hybrid Radial Basis Function Neural Network (RBFNN) and Kolmogorov-Arnold Network (KAN) model, termed RBF-KAN, for accurate indoor localization using RSSI-based fingerprinting. The combination leverages RBFNN and KAN for handling complex patterns adaptively. Our experimental results demonstrate that RBF-KAN achieves superior performance, with an average error of 0.105m, an accuracy of 98.26 %, and fast convergence, outperforming other methods like KAN, Deep-KAN, RBFNN, MLP, and RBF-MLP.