An Explainable Artificial Intelligence empowered energy efficient indoor localization framework for smart buildings

Abstract

The indoor localization problem remains a prominent and extensively debated area of research, lacking a universally accepted solution, especially within the context of smart buildings. A major concern revolves around the energy consumption associated with indoor localization systems. This study presents a proposed framework for an energy-efficient indoor localization system designed for smart buildings. The approach focuses on a fingerprinting indoor localization technique that involves constructing a signal map. To address challenges arising from distinct signal effects and the environment-specific structure of signal maps, the study introduces a framework incorporating an adaptive filter selection scheme. This scheme includes Kalman, particle, and Savitzky-Golay filters in the pre-processing stage to enhance the signal map. Rather than resorting to additional hardware for improved localization accuracy, the study advocates for optimizing the signal map to minimize energy consumption. Additionally, the research emphasizes the selection of effective features for machine learning techniques to enhance performance and boost localization accuracy. The findings are subjected to analysis using Interpretable Model-agnostic Explanations (LIME) and Shapley Additive exPlanations (SHAP) Explainable Artificial Intelligence (XAI) models. The investigation delves into the impact of each signal and filter on positioning estimation, providing a comprehensive understanding of the system's functionality.