I-SEA: Improved shape exchange algorithm for quasi-periodic time series alignment

Abstract

Dynamic Time Warping (DTW) is one of the most important algorithms for time series alignment. However, it is unsuitable for quasi-periodic time series. These are a concatenation of quasi-similar forms called (quasi)periods, e.g. electrocardiogram (ECG) time series. It is even more difficult to align these series when each contains a different number of periods. The difficulty lies in the fact that each period is characterized by local morphological changes. SEA (Shape Exchange Algorithm) is the only algorithm that effectively aligns these very complex time series. However when it comes to aligning time series significantly phase shifted and contaminated with noise, the SEA algorithm shows some weaknesses. To remedy to this problem, we propose in this work a novel algorithm that combines the SEA and DTW algorithms. The new method is then called I-SEA (Improved shape Exchange Algorithm). The tests were performed on ECG time series, selected from the Massachusetts Institute of Technology - Beth Israel Hospital (MIT-BIH) public database. Results show that the proposed algorithm is more effective than SEA in terms of alignment accuracy on both qualitative and quantitative levels, which makes it very suitable for many applications related to time series data mining (searching, classification, diagnosis, etc.), for many types of media.