Research of Detection Algorithm for Time Series Abnormal Subsequence

The recent advancements in sensor technology have made it possible to collect enormous amounts of data in real time. How to find out unusual pattern from time series data plays a very important role in data mining. In this paper, we focus on the abnormal subsequence detection. The original definition of discord subsequences is defective for some kind of time series, in this paper we give a more robust definition which is based on the k nearest neighbors. We also donate a novel method for time series representation, it has better performance than traditional methods (like PAA/SAX) to represent the characteristic of some special time series. To speed up the process of abnormal subsequence detection, we used the clustering method to optimize the outer loop ordering and early abandon subsequence which is impossible to be abnormal. The experiment results validate that the algorithm is correct and has a high efficiency.