An intelligent method of change-point detection based on LS-SVM algorithm

Abstract

The change-point detection problem is an important consideration in many application areas. This article aims to address this issue in auto-correlated process, where traditional statistical methods are not applicable. Based on least squares-support vector machine (LS-SVM) pattern recogniser, this article develops an intelligent method for solving the problem of the change-point detection, and the proposed model is applied to detect the change-point of process mean-shift in auto-correlated time-series process. In this research, LS-SVM algorithm and moving window method are used to detect the location of the mean-shift signal. The LS-SVM pattern recogniser is designed and the performance of the recogniser is evaluated in terms of Accuracy Rate. Results of simulation experiment show that the proposed intelligent model is an effective method to detect the change-point in the mean of autoregressive moving average data series. Compared with the traditional statistical methods for the change-points detection, the proposed intelligent method can work well without the need for the prior information of the process data. Though the proposed method in this article is only used in solving the problem of single change-point detection in the mean, it can be potentially extended and used in solving the practical problems of multiple change-point detections.

Keywords:

• change-point detection
• least squares-support vector machine
• moving window
• pattern recognition
• autoregressive moving average process

Acknowledgement

This work was supported by Humanities and Social Sciences Planning Fund of China's Ministry of Education (12YJA630030) and National Natural Science Foundation of China (70931002).