A new algorithm in detecting changepoint in linear regression models

Abstract

The changepoint problem in a two-phase simple linear regression model has received increasing attentions. A changepoint in copy number variations in bioinformatics and medical informatics, pattern recognitions, data mining, and many other applications, refers to a time point at which a structural pattern change occurs during a long-term experimentation process. Given a series of observations, the problem is to detect a putative changepoint in the series. Computations in detecting a changepoint is typically time-consuming and inefficient. Recently Liu and Qian (2010) proposed an interesting and computationally easy algorithm via empirical likelihood methods. In this article, a new algorithm is proposed to improve the detecting power. The new algorithm is computationally as easy as Liu and Qian's algorithm. Simulation results show that the new algorithm greatly improves the detecting powers and hit rates over Liu and Qian's algorithm.