Online Time Series Segmentation Using Temporal Mixture Models and Bayesian Model Selection

This paper is concerned with the issue of online time series segmentation. This problem, common in a number of applicative fields, continues to receive increasing attention. The present article introduces a novel threshold-free sequential time series segmentation approach. It is based on the concurrent estimation of two models (a model with one regressive segment and a two-component temporal mixture model adapted to the time series segmentation framework) and uses the Bayesian Information Criterion to decide between the two models. The proposed approach is shown to be efficient using a variety of simulated time series and a real-world time series arising from a railway application.