Performance Comparison of the Longest Common Subsequence and Dynamic Time Warping in time series data mining

Today, the use of various data mining techniques such as Classification, clustering, rule discovery, Query by content, forecasting in different domains including Production, medicine, social, meteorology, stock exchange, sales, customer service and other areas are increasing. But, these techniques are specially designed for fixed data, so using them for time series data requires some changes in the corresponding algorithms like select a suitable similarity measurement. According to the recent study, the Longest Common Subsequence (LCSS) and Dynamic Time Warping (DTW) methods are the most widely used and effective methods for measuring similarity in time series data mining. In this research, the accuracy of these methods in nearest neighbor and K-medoids clustering techniques on 63 datasets from the UCR collection will be examined and compare with the pairwise comparison test. Based on the results, these methods are differ significantly in terms of accuracy in determining the correct class of time series with nearest neighbor technique, but they do not differ significantly in terms of accuracy detect the cluster representative and cluster number with K-medoids technique. However, it s still important to note that the performance of these methods is different and somewhat weak in relation to some of the datasets..

کلیدواژه ها:

Time Series Data mining ، Performance ، Similarity ، Clustering ، nearest neighbor ، Longest Common subsequence ، Dynamic Time Warping