Change-Point Detection with Local Trend Adjustment

Identifying the number and precise locations of multiple change points in long sequences is a critical issue in statistics and machine learning.However, accurate change point detection can be compromised by the presence of local trends in the sequence when using the conventional parametric piecewiseconstant model.In this paper, we introduce an adaptive Neyman test to assess the presence of local trends.Subsequently, we develop a novel change point detection procedure based on a partially linear model that incorporates these local trends.Furthermore, we extend the proposed testing and estimation methods to multidimensional cases, facilitating the identification of common change points in array-based data.Our methods are straightforward to implement, and we evaluate their numerical performance through simulations and the analysis of SNP genotyping data.