Efficient Mining of Density-Aware Distinguishing Sequential Patterns with Gap Constraints

Distinguishing sequential patterns are useful in characterizing a given sequence class and contrasting that class against other sequence classes. This paper introduces the density concept into distinguishing sequential pattern mining, extending previous studies which considered gap and support constraints. Density is concerned with the number of times of given patterns occur in individual sequences; it is an important factor in many applications including biology, healthcare and financial analysis. We present gd-DSPMiner, a mining method with various pruning techniques, for mining density-aware distinguishing sequential patterns that satisfy density and gap, as well as support, constraints. With respect to computational speed, when the procedures related to density are masked gd-DSPMiner is substantially faster than previous distinguishing sequential pattern mining methods. Experiments on real data sets confirmed the effectiveness and efficiency of gd-DSPMiner in the general setting and the ability of gd-DSPMiner to discover density-aware distinguishing sequential patterns.