Bayesian Changepoints Detection for the Power Law Process with Binary Segmentation Procedures

Abstract

We consider the power law process which is assumed to have multiple changepoints. We propose a binary segmentation procedure for locating all existing changepoints. We select one model betwen the no-changepoints model and the single changepoint model by the Bayes factor. We repeat this procedure until no more changepoints are found. Then we carry out a multiple test based on the Bayes factor through the intrinsic priors of Berger and Pericchi (1996) to investigate system behaviour of failure times. We demonstrate our procedure with a real dataset and some simulated datasets.