An extensible framework for ECG anomaly detection in wireless body sensor monitoring systems

Abstract

Automatic anomaly detection from the sensor streams of the Electrocardiography requires online techniques of data stream processing and analysis. The significant increase in the availability of the Electrocardiography collected from wireless sensor networks (WSNs) has attracted heaps of research interests in identifying the anomalies in an Electrocardiography. However, most existing models focus on a particular type of anomaly, like arrhythmia, and they are not dynamically extensible for the identification of an unknown type of anomaly. This work proposes an extensible method named shapelet-base (SH-BASE) to solve this problem. Essentially, SH-BASE is a knowledge-base of shapelets, in which the shapelets are classified into different groups (we call them branches), and each branch stores the most distinctive shapelets of a particular type of anomaly. Experimental results show that the SH-BASE can achieve a very competitive performance in anomaly detection for Electrocardiography sensor streams by comparing with the state-of-the-art models.