Speeding Up Similarity Search on a Large Time Series Dataset under Time Warping Distance

Time series data are a ubiquitous data type appearing in many domains such as statistics, finance, multimedia, etc. Similarity search and measurement on time series data are typically different from on other data types since time series data have the associations among adjacent dimensions. Accordingly, the classic Euclidean distance metric is not an accurate similarity measure for lime series. Therefore, Dynamic Time Warping (DTW) has become a better choice for similarity measurement on time series in various applications regardless of its high computational cost. To speed tip the calculation, many research works attempt to speed up DTW calculation using indexing method, which always has a tradeoff between indexing efficiency and I/O cost. In this paper, we propose a novel method to balance this tradeoff under indexed sequential access using Sequentially Indexed Structure (SIS), an approach to time series indexing with low computational cost and small overheads on I/O. Finally, we conduct experiments to demonstrate our superiority in speed performance over the best existing method.