Overlapping trend detection and application in prediction
Xuedong Gao, Gu Kan, Danyue Wang · 2016
Traditional time series studies pay more attention to the representation effect of original sequence rather than to the trend contained in it. And overlapping trends are ignored because of the usual way for extracting trend from time series is decomposition or segmentation. A sufficient long sequence is also required when doing prediction with trend analysis, or only disappointing results would be gotten. This paper introduces sliding window to improve the inertia test based trend detection algorithm. The data of experiments show that the improved algorithm gets the target of overlapping trend detection and short sequence prediction.