Subgroup Discovery in Sequential Databases

Sequential pattern mining produces a vast number of frequent patterns due to the combinatorial nature of the problem and redundant information in the results. Several pattern mining techniques (e.g., closed-patterns, maximal-patterns, gap constraints, recency and compactness constraints) have been studied to reduce the size of the results. However, these approaches can still generate large result sets often containing patterns of little use or interestingness to the end users. Even when many interesting results are returned, finding the most useful can be difficult. By applying ideas from subgroup discovery to sequential pattern mining, we have developed exact and heuristic-based algorithms for identifying and ranking the top-k most significant patterns from the complete collection of frequent patterns.