Sequential pattern mining is an efficient technique for discovering recurring structures or patterns from very large datasets, with a very large field of applications. It aims at extracting a set of attributes, shared across time among a large number of objects in a given database. Previous studies have developed two major classes of sequential pattern mining methods, namely, the candidate generation-and-test approach based on either vertical or horizontal data formats represented respectively by GSP and SPADE, and the pattern-growth approach represented by FreeSpan, PrefixSpan and their further extensions. The performances of these algorithms depend on how patterns grow. Because of this, we introduce a heuristic to predict the optimal pattern-growth direction, i.e. the pattern-growth direction leading to the best performance in terms of runtime and memory usage. Then, we perform a number of experimentations on both real-life and synthetic datasets to test the heuristic. The performance analysis of these experimentations show that the heuristic prediction is reliable in general.
A pattern growth-based sequential pattern mining algorithm called prefixSuffixSpan
