Recognizing Emergencies and Multi-User Behavior Patterns Using Imperfect Data From Distributed Access Points. A Non-Intrusive Proof of Concept

Abstract

This paper presents a privacy-preserving proof of concept for assessing human behavior in emergency scenarios using aggregated data from multiple WiFi access points. The proposed method focuses on preserving individual privacy by avoiding tracking and metadata analysis, while still achieving effective multi-user activity recognition. To implement our approach, raw data from the Eduroam WiFi network at Polytechnique Montreal was collected and analyzed using standard supervised and anomaly detection techniques. The initial test was on recognizing patterns of academic activity, serving as the foundation for our investigation. Subsequently, the same methodology was applied during an evacuation drill scenario to recognize anomaly situations. Through our research, we demonstrate the potential to assess human situations effectively while safeguarding privacy, providing a critical capability for the early detection of emergency situations.

Keywords

• Wireless networks
• movement patterns
• indoor behavior
• machine learning
• binary classification
• anomaly detection