Learning model trees from evolving data streams
Data Mining and Knowledge Discovery
Change detection with Kalman filter and CUSUM
Ubiquitous knowledge discovery
Change detection with Kalman filter and CUSUM
Ubiquitous knowledge discovery
Change detection with kalman filter and CUSUM
DS'06 Proceedings of the 9th international conference on Discovery Science
Hi-index | 0.00 |
Structural break is one of the important concerns in non-stationary time series prediction The cumulative sum of square (CUSUMS) statistic proposed by Brown et al (1975) has been developed as a general method for detecting a structural break To better utilize this method, this paper analyses the operating conditions of the centered version of CUSUMS using three variables: the percentage of variance change, the post-break data size and the pre-break data size In traditional approach of the centered CUSUMS, all available data are used for the break detection Our analysis reveals that one can improve the accuracy of the break detection by either reducing the post-break data size or increasing pre-break data size Based on our analysis, we propose a modified test statistic The evidence shows that the modified statistic significantly improves the chance of detecting the structural breaks.