Alignment of Time Series for Subsequence-to-Subsequence Time Series Matching

Abstract

The success of time series data mining applications, such as query by content, clustering, and classification, is greatly determined by the performance of the algorithm used for the determination of similarity between two time series. The previous research on time series matching has mainly focused on whole sequence matching and to limited extent on sequence-to-subsequence matching. Relatively, very little work has been done on subsequence-to-subsequence matching, where two time series are considered similar if they contain similar subsequences or patterns in the same time order. This paper presents an effective approach capable of handling whole sequence, sequence-to-subsequence and subsequence-to-subsequence matching. The proposed approach derives its strength from the novel two stage segmentation algorithm, which facilitates the alignment of the two time series by retaining perceptually important points of the two time series as break points.