XDOcc: An Explainable Artificial Intelligence Empowered Deep Framework for Occupancy Detection and Occupant Count Estimation

Abstract

This study introduces explainable deep occupancy, a robust framework empowered by explainable artificial intelligence for high-accuracy occupancy detection and occupant count estimation in smart buildings. Despite the complexity of indoor environments, the proposed framework demonstrates strong performance without requiring filtering or data balancing in the pre-processing phase, utilizing raw signal data from environmental sensors and WiFi access points. The deep learning architecture is based on recurrent neural network models, specifically gated recurrent unit and long short-term memory, to capture temporal dependencies. Gated recurrent unit, long short-term memory, bidirectional gated recurrent unit, and bidirectional long short-term memory models were implemented, along with two hybrid models combining bidirectional gated recurrent unit and bidirectional long short-term memory. The first hybrid merges outputs of both models to enhance representational power. The second model integrates an attention mechanism to focus on critical temporal patterns. All models were evaluated on four datasets collected from three distinct buildings. The hybrid model with the attention mechanism achieved the best performance and was selected for integration into the final framework. To ensure interpretability, the results were analyzed using the Shapley Additive Explanations method, enabling identification of key hardware components and important sensor features contributing to prediction accuracy. Moreover, redundant components with minimal impact were identified, supporting hardware optimization and energy efficiency. The proposed framework achieved a maximum accuracy of 0.999703 for occupancy detection and 0.997917 for occupant count estimation, each obtained on different benchmark datasets. Obtained results demonstrate the framework's strong generalization capability across diverse indoor environments.