Similar Time Series Retrieval Using Only Important Segments

Similar time series searching plays an important role in applications such as time series classification and outlier detection. We observe that different segment of a time series may have different significance, thus propose to assign different weight to each segment, and extract those segments with highest weights for distance computation. Since these segments are more representative, we can achieve high accuracy of similarity search with much lower computation overhead. The result of experiments on both real world and synthetic data sets demonstrates that we can achieve comparable or even higher accuracy while largely reduce the computation overhead, if we use only those important segments rather than the whole time series while performing similarity search.