xKAN: Explainable AI for Smartphone-Based Indoor Localization Using Wi-Fi Fingerprinting
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Machine learning (ML) models have become central to Wi-Fi based indoor localization, where noisy and highly variable fingerprints must be mapped to physical locations. Despite their strong performance, state-of-the-art ML models typically operate as opaque black boxes, offering limited visibility into the factors driving their predictions. This lack of intrinsic interpretability becomes even more critical in real indoor environments, where Wi-Fi data is inherently non-independent and non-identically distributed (non-IID) due to device heterogeneity, environmental dynamics, and floorplan-specific variations. Under these conditions, trust, reliability, and diagnosability remain major barriers to deploying ML-based indoor localization systems in practice. To address these challenges, we present xKAN, a post-hoc, model-agnostic explainability framework designed to improve the transparency of ML models for indoor localization. xKAN provides two complementary forms of explainability: global explainability, which summarizes how a model leverages Wi-Fi signals across an entire floorplan, and local explainability, which highlights the specific signal factors influencing an individual fingerprint’s prediction. Together, these perspectives reveal how broader Wi-Fi signal patterns shape local decision outcomes, uncovering hidden biases, inconsistencies, and region-specific behaviors within the ML model’s decision process. Experimental results demonstrate that xKAN offers a structured and intuitive pathway toward explainable AI driven indoor localization, with most relevant first (MoRF) evaluations showing that xKAN achieves up to 1.10× - 1.43× over SHAP and 1.03× - 1.38× over LIME in terms of faithfulness compared to state-of-the-art explainability methods. By improving transparency without modifying the underlying model, xKAN supports the development of more robust, reliable, and deployable localization systems for real-world environments.
Fields of Study & Subjects
Explainable Artificial Intelligence (XAI)
Indoor and Outdoor Localization Technologies
Optical Wireless Communication Technologies
Publication details
- Journal / source
- IEEE Access
- Volume
- 14
- Pages
- 49550–49562
- Publication date
- 1 January 2026
- Language
- EN