General algorithms for mining closed flexible patterns under various equivalence relations

Abstract

We address the closed pattern discovery problem in sequential databases for the class of flexible patterns. We propose two techniques of coarsening existing equivalence relations on the set of patterns to obtain new equivalence relations. Our new algorithm GenCloFlex is a generalization of MaxFlex proposed by Arimura and Uno (2007) that was designed for a particular equivalence relation. GenCloFlex can cope with existing, as well as new equivalence relations, and we investigate the computational complexities of the algorithm for respective equivalence relations. Then, we present an improved algorithm GenCloFlex+ based on new pruning techniques, which improve the delay time per output for some of the equivalence relations. By computational experiments on synthetic data, we show that most of the redundancies in the mined patterns are removed using the proposed equivalence relations. © 2012 Springer-Verlag.