Unraveling Sensor Correlations in Multi-Sensor Wearable Devices for Smart Anomaly Detection

In health monitoring and activity tracking technologies, wearable or implantable sensors have become indispensable, linking various human body regions to collect vital health data. Despite their potential, ensuring the security and reliability of these devices presents significant challenges, primarily due to the complexity of real-world scenarios that these systems encounter. Current approaches often rely on anomaly detection models that process historical sensor data to identify issues. However, these models tend to falter when faced with unexpected conditions or 'corner cases,' lacking the ability to generalize across the diverse situations encountered in everyday use. This limitation is particularly critical in wearable devices, where unexpected incidents are of paramount importance and cannot be overlooked. Addressing this gap, our research investigates multi-sensor wearable systems to understand the context of system operations and their characteristics. We introduce a context-aware approach that leverages the unique physics of the human body to identify the intricate relationships between sensors. By extracting sensor relations and patterns, our approach aims to enhance the detection of security and reliability issues, offering an advancement over traditional methods.