Efficient data-driven occupancy detection in office environments and feature impact analysis

Occupancy detection is crucial in optimizing building energy efficiency and enhancing occupant comfort. This study introduces an innovative data-driven approach for accurate occupancy detection in an office room environment. Specifically, the methodology combines the advantages of Independent Component Analysis (ICA) to extract essential features from multivariate data and Kantorovitch distance (KD)-based schemes for detection sensitivity. The KD scheme’s detection threshold is computed nonparametrically using kernel density estimation to enhance the sensitivity of occupancy detection. The efficacy of this strategy is evaluated utilizing publicly available data recorded during winter in Mons, Belgium, capturing vital environmental parameters such as temperature, humidity, light, and CO \(_{2}\) $_2^{}$ levels through specialized sensors. Results demonstrate that the ICA-KD approach achieves an averaged accuracy of 98.355%, surpassing conventional approaches like Principal Component Analysis (PCA)-based, ICA-based, and other state-of-the-art methods. Additionally, the study uses Shapley Additive exPlanations (SHAP) with XGBoost to explore the impact of input variables on occupancy detection, highlighting the influence of various factors under different testing conditions.