Elastic Non-contiguous Sequence Pattern Detection for Data Stream Monitoring

Abstract

In recent years, there has been an increasing interest in the detection of non-contiguous sequence patterns in data streams. Existing works define a fixed temporal constraint between every pair of adjacent elements of the sequence. While this method is simple and intuitive, it suffers from the following shortcomings: 1)It is difficult for the users who are not domain experts to specify such complex temporal constraints properly; 2)The fixed temporal constraint is not flexible to capture interested patterns hidden in long sequences. In this paper, we introduce a novel type of non-contiguous sequence pattern, named Elastic Temporal Constrained Non-contiguous Sequence Pattern(ETC-NSP). Such a pattern defines an elastic temporal constraint on the sequence, thus is more flexible and effective as opposed to the fixed temporal constraints. Detection of ETC-NSP in data streams is a non-trivial task since a brute force approach is exponential in time. Our method exploits an similarity measurement called Minimal Variance Matching as the basic matching mechanism. To further speed up the monitoring process, we develop pruning strategies which make it practical to use ETC-NSP in streaming environment. Experimental studies show that the proposed method is efficient and effective in detecting non-contiguous sequence patterns from data streams.