AI-based Occupancy Prediction using WiFi CSI

Occupancy detection is crucial for efficiently managing energy in building. However, purchasing occupancy sensors incurs additional cost, and their accuracy is typically low. This paper proposes AI-based occupancy prediction model using Channel State Information (CSI) from WiFi APs. Two models have been implemented based on logistic regression and deep neural network. By synergistically combining their respective advantages, its accuracy can be improved. This enables high-accuracy occupancy prediction using existing WiFi APs without deploying new sensors. Experimental results show that proposed model achieves a high accuracy exceeding 94%, which is an acceptable rate for real-world applications.