The objective of change-point detection is to discover abrupt propertychanges lying behind time-series data. In this paper, we present a novelstatistical change-point detection algorithm based on non-parametric divergenceestimation between time-series samples from two retrospective segments. Ourmethod uses the relative Pearson divergence as a divergence measure, and it isaccurately and efficiently estimated by a method of direct density-ratioestimation. Through experiments on artificial and real-world datasets includinghuman-activity sensing, speech, and Twitter messages, we demonstrate theusefulness of the proposed method.
展开▼
