Wi-SafeHome: WiFi Sensing Based Suspicious Activity Detection for Safe Home Environment

Recently, WiFi signals are being used for sensing task based applications in addition to standard communication activities. Specifically, the Channel State Information (CSI) extracted from WiFi signals through channel estimation at the receiver end provides unique information about environmental dynamics. This CSI data is used for various tasks including motion and human presence detection, localization, environmental monitoring, and a few other sensing applications. By analyzing both the amplitude and phase of the CSI data, we can gain intricate insights into how signal transmission paths are affected by physical and environmental changes. In this study, we focus on leveraging low-cost WiFi-enabled ESP32 micro-controller devices to monitor suspicious-related activities within indoor environments. We conducted exhaustive experiments involving four distinct suspicious-related human activities leaving a room, entering a room, sneaking into a room without formal entry, and engaging in suspicious activities within a room. To enhance the detection rate for these activities, we employ feature engineering techniques on the received CSI data. Additionally, we applied a low-pass filter to eliminate noise from the received signal effectively. To achieve accurate suspicious activity classification, we harnessed various lightweight machine learning (ML) algorithms, which include Support Vector Machine (SVM), Random Forest (RF), Gradient Boosting, Extreme Gradient Boosting (XG Boost), and K-Nearest Neighbor (KNN). Our results reveal that KNN outperformed the other ML models, achieving an accuracy rate of 99.1% and F1-Score of 0.99. This suggests that KNN is a robust choice for effectively classifying suspicious activities based on WiFi CSI data.