一个基于主成分分析的探测高维变点的方法

An Approach to Detect High-dimensional Change-points Based on Principle Component Analysis

本文介绍了一个有效的处理高维变点问题的方法。我们先将数据矩阵使用主成分分析的方法投影到低维空间,然后再利用传统变点的方法来进行估计。在变点个数未知时,我们使用交叉核实的方法来估计变点个数。在数值模拟研究中,我们将新方法同一些已有的方法进行了比较,在估计的准确度和计算时间等方面都要优于其他方法。

In this paper,we propose an efficient method for detecting change-points in high-dimensional setting.We first utilize the principle component analysis to project the data matrix to a lower-dimensional space,and then estimate the location of change-points by conventional change-points detection method.When the number of change-points is unknown,we use the cross-validation to estimate the number.In the simulation studies,we compare the new method with some existing methods.Our method performs well in both aspects of estimation accuracy and computational time.