RECOGNIZING CHARACTERISTIC PATTERNS IN DISTORTED DATA COLLECTIONS

摘要

Many models and artificial intelligence methods work with the inputs in the form of time series. Generally, success of many of them strongly depends on ability to successfully manage input data, which often contains repeating similar episodes (patterns). If these patterns are recognized, they can be used for instance for indexing, prediction or compression. These operations can also be very useful for improving the already existing model performance and accuracy. Our effort is to provide a robust mechanism for retrieving these characteristic patterns from the collections that are subject of various distortions. The whole process of our pattern recognition consists of receiving ,the episodes, their clustering into the groups of similar episodes and deriving the representatives of each cluster. These representatives will be used for further indexing collections. This paper is focused on the last step of this process - receiving the representatives of concrete clusters using.Dynamic Time Warping method.