Subseries Join: A Similarity-Based Time Series Match Approach

Time series data appears in numerous applications including medical data processing, financial analytics, network traffic monitoring, and Web click-stream analysis. An essential task in time series mining is efficiently finding matches between similar time series or parts of time series in a large dataset. In this work, we introduce a new definition of subseries join as a generalization of subseries matching. We then propose an efficient and robust solution to subseries join (and match) based on a non-uniform segmentation and a hierarchical feature representation. Experiments demonstrate the effectiveness of our approach and also show that this approach can better tolerate noise and phase-scaling than previous work.