Searching Time Series Based On Pattern Extraction Using Dynamic Time Warping

Many types of data collections processed by time series analysis often contain repeating similar episodes (patterns). If these patterns are recognized, then they may be used for instance in data compression, for prediction or for indexing large collections. Extraction of these patterns from data collections with components generated in equidistant time and in finite number of levels is now a trivial task. The problem arises for data collections that are a subject to different types of distortions in all axes. In this type of collections, the found similar episodes do not have to be exactly the same; they can differ in time, shape or amplitude. In these cases, it is necessary to pick the suitable one from each group of similar episodes and to declare it as a representative member of the whole group. This paper discusses the possibilities of using the Dynamic Time Warping (DTW) method for deriving the representative member of a group of similar episodes that are subjects to the previously mentioned distortions. The paper is also focused on providing a suitable mechanism for more effective searching of distorted time series.