Fall Detection using Random Forest under Wi-Fi Interference

Sammanfattning

Falling accidents represent one of the most common causes of injury in elderly care, which makes it necessary to have non-invasive monitoring systems that support independent living while preserving privacy. To address this, the thesis presents a device-free and non-invasive fall detection system using Integrated Sensing and Communication (ISAC) to detect fall accidents. The system uses a network of ESP32-C3 microcontrollers to collect Fine Time Measurement (FTM), Received Signal Strength Indicator (RSSI), and Channel State Information (CSI). These metrics are processed by a Random Forest classifier to distinguish between falls and normal activities. Additionally, the project focuses on the Taguchi method to evaluate and optimize the balance between Wi-Fi sensing and communication performance. The results demonstrate that Wi-Fi interference has an impact on the system's False Positive Rate (FPR); it decreases network download speed by 23.43% and increases network latency by 15.79%. Under optimal settings within a static environment and with minimal network interference, the system achieved an F1-score of 76.19% while the download speed decreased by only 5.85%. Finally, this work contributes to balancing communication performance with Wi-Fi sensing. The thesis shows that the ISAC can successfully detect falls without relying on cameras or requiring users to wear devices, while ensuring a limited impact on overall network performance.