SAXJS: An Online Change Point Detection for Wearable Sensor Data

Abstract

Wearable electronics are devices used by humans that can continuously and uninterruptedly monitor human activity through sensor data. The data collected by them have several applications, such as recommending running techniques and helping to monitor health status. Segmenting such data into chunks containing only a single human activity is challenging due to the wide variability of underlying process characteristics presented in the data. To deal with this problem, we propose a new change point detection algorithm based on the Symbolic Aggregate approXimation (SAX) transformation, the probability of transition between symbols, and the Jensen-Shannon distance.