Paper detail

A Novel Approach for Fast Detection of Multiple Change Points in Linear Models

A change point problem occurs in many statistical applications. If there exist change points in a model, it is harmful to make a statistical analysis without any consideration of the existence of the change points and the results derived from such an analysis may be misleading. There are rich literatures on change point detection. Although many methods have been proposed for detecting multiple change points, using these methods to find multiple change points in a large sample seems not feasible. In this article, a connection between multiple change point detection and variable selection through a proper segmentation of data sequence is established, and a novel approach is proposed to tackle multiple change point detection problem via the following two key steps: (1) apply the recent advances in consistent variable selection methods such as SCAD, adaptive LASSO and MCP to detect change points; (2) employ a refine procedure to improve the accuracy of change point estimation. Five algorithms are hence proposed, which can detect change points with much less time and more accuracy compared to those in literature. In addition, an optimal segmentation algorithm based on residual sum of squares is given. Our simulation study shows that the proposed algorithms are computationally efficient with improved change point estimation accuracy. The new approach is readily generalized to detect multiple change points in other models such as generalized linear models and nonparametric models.