Real-time Spatio-Temporal based Outlier Detection Framework for Wireless Body Sensor Networks

Abstract

Wireless body sensor networks (WBSNs) are threatened by many issues like anomalies in collected data and failure in their hardware components. An outlier detection approach applied on online monitoring of vital signs can both prevent collection of outlier data and detect emergent health degradation. In this paper, we propose an outlier detection framework for real time sensed data by WBSNs. Our proposed solution is twofold: Robust z score algorithm is executed at first step on the sensor nodes level to detect abnormal values and send them to the coordinator. After that, Isolation Forest is executed at the coordinator to distinguish between a faulty measurement and a critical health state. Correlation among vital signs are exploited to differentiate between an emergent healthy event and an anomaly in the measured data. Experiments conducted on real physiological datasets show that our proposed method is able to achieve a good detection accuracy with a low false alarm rate. Complexity and energy efficiency studies demonstrate the low complexity and lightness of our proposed solution.