Eecient Time-series Subsequence Matching Using Duality in Constructing Windows Eecient Time-series Subsequence Matching Using Duality in Constructing Windows

Subsequence matching in time-series databases is an important problem in data mining and has attracted a lot of research interest. It is a problem of nding the data sequences containing subsequences similar to a given query sequence and of nding the oosets of these subsequences in the original data sequences. In this paper, we propose a new approach (Dual Match) to subsequence matching that exploits duality in constructing windows and show this approach signiicantly improves performance. Dual Match divides data sequences into disjoint windows and the query sequence into sliding windows, and thus, is a dual approach of the one by Faloutsos et al. (FRM in short), which divides data sequences into sliding windows and the query sequence into disjoint windows. FRM causes a lot of false alarms (i.e., candidates that do not qualify) by storing minimum bounding rectangles rather than individual points representing windows to save storage space for the index. Dual Match solves this problem by directly storing points without incurring excessive storage overhead. Experimental results show that, in most cases, Dual Match provides large improvement both in false alarms and performance over FRM given the same amount of storage space. In particular, for low selectivities (less than 10 ?4), Dual Match drastically reduces the number of candidates|down to as little as 1 8800 of that for FRM|reduces the number of page accesses by up to 26.9 times, and improves performance up to 430-fold. On the other hand, for high selectivities (more than 10 ?2), it shows a very minor degradation (less than 29%) by all three measures. For selectivities in between (10 ?4 10 ?2), Dual Match shows performance slightly better than that of FRM. Dual Match also provides excellent performance in index creation. Experimental results show that it is 4.1025.6 times faster than FRM in building indexes of approximately same sizes. The main reason is that it requires far smaller number of transformations, which are a major part of the CPU overhead, than FRM does. Overall, these results indicate that our approach provides a new paradigm in subsequence matching that improves performance signiicantly in large database applications.