FastSEA: A very fast and very effective matching technique for very complex time series

Abstract

Complex time series are a concatenation of quasi-periodic time series, e.g. electrocardiogram (ECG) and capnogram. The complexity lies in the fact that these series usually composed of different number of periods, of different lengths that might be shifted/scaled on the time/magnitude axis. Due to this complexity, time series matching is rendered more difficult. The Shape Exchange Algorithm (SEA) and its derivates are currently the only algorithms that can effectively match this type of time series. However, SEA usefulness is limited only to small time series due to its quadratic temporal complexity. In this paper, an accelerated version of the SEA method is proposed that has a linear computational complexity. The new method is referred to by the acronym FastSEA. It is based on more efficient and yet simple sorting algorithm that is Counting Sort Algorithm. We have evaluated the new developed FastSEA method on ECG signals, selected from (MIT-BIH) public database. The experimental results show that FastSEA outperforms SEA in terms of efficiency by time reduction up to 99 % without loss with respect to SEA in terms of effectiveness. Also, FastSEA is found to be comparable to previously published speed up techniques for SEA: ASEAL and FANSEA, besides it obviously being more interesting than the famous DTW method, with no loss in accuracy. As a result, the proposed FastSEA is able to align very long time series in very short periods of time. Needless is to mention that FastSEA inherits the ability of SEA of being suitable for many applications, such as time series diagnosis and person identification using ECG, etc.