Improving Frequent Patterns Mining by LFP

Frequent patterns mining is the focused research topic in association rule analysis. Most of the previous studies adopt Apriori-like algorithms or lattice-theoretic approaches which generate-and-test candidates. However, there are extremely invalidated candidate generations in the exponential search space. In this paper, we systematically explore the search space of frequent patterns mining and present a local frequent pruning (LFP) strategy based on local frequent property. LFP can be used in all Apriori-like algorithms. With a little more memory overhead, proposed pruning strategy can prune invalidated search space and effectively decrease the total number of infrequent candidate generation. For effectiveness testing reason, we optimize MAFIA and SPAM and present the improved algorithms, MAFIA+ and SPAM+. A comprehensive performance experiments study shows that LFP can improve performance by a factor of 10 on small datasets and better than 30% to 50% on reasonably large datasets.