Discover Motifs in Multi-dimensional Time-Series Using the Principal Component Analysis and the MDL Principle

Recently, the detection of previously unknown, frequently occurring pattern is regarded as a difficult problem. We call this pattern as “motif”. Many researchers have proposed algorithms for discovering the motif. However, if the optimal period length of the motif is not known in advance, we can not use their algorithms for discovering the motif. In this paper, we attempt to determine the optimum period length dynamically using the MDL principle. Moreover, in order to apply this algorithm to the multi dimensional time-series, we transform the time-series into one dimensional time-series by using the Principal Component Analysis. Finally, we show experimental results and discuss the efficiency of our motif discovery algorithm.