Font Size: 

Changepoint detection aims to find abrupt changes in time series data. These changes denote substantial modifications to the process; they can vary from simple changes in location to a change in distribution. Traditional changepoint detection methods often rely on a cost function to assess if a change occurred in a series. Here, changepoint detection in a clustering framework is investigated, and a novel changepoint detection algorithm is developed using a finite mixture of regressions with concomitant variables. Through the introduction of a label correction mechanism, the unstructured cluster labels are treated as ordered and distinct segment labels. Different kinds of change can be captured using a parsimonious family of models. Performance is illustrated on simulated and real data.

mixture models, changepoint detection, family of models