Mining Top-K Distinguishing Temporal Sequential Patterns From Event Sequences

Sequential patterns are useful in many areas such as biomedical sequence analysis, web browsing log analysis, and historical banking transaction log analysis. Distinguishing sequential patterns can help characterize the differences between two or more sets/classes of sequences, and can be used to understand those sequence sets/classes and to identify informative features for classification and so on. However, previous studies have not considered how to mine distinguishing sequential patterns from event sequences, where each event in a sequence has an associated timestamp. To fill that gap, this paper considers the mining of distinguishing temporal event patterns (DTEP) from event sequences. After discussing the challenges on DTEP mining, we present DTEP-Miner, a mining method with various pruning techniques, for mining DTEPs with top-k contrast scores. Our empirical study using both real data and synthetic data demonstrates that DTEP-Miner is effective and efficient.