Maximal fuzzy supplement frequent pattern mining based on advanced pattern-aware dynamic search strategy and an effective FSFP-array technique

Abstract

The proper expression of the potentially useful but hidden information in large-scale datasets via using proper structure is vital important in both theory and applications of advanced pattern mining. The fundamental challenges are how to alleviate the mining combinatorial explosion problem and ensure the efficiency of mining results. However, most of the existing algorithms have not been entirely capable of solving these issues due to the fact that enormous number of candidate patterns has been generated and the weight constraints of items were only considered in crisp values. In order to generate more practical patterns in the new proposed Fuzzy Supplement Frequent Pattern (FSFP), base-(second-order-effect) pattern structure is proposed and new pruning strategies including pattern-aware dynamic base pattern search strategy and FSFP-array technique are given. Thus, the proposed maximal FSFPs mining algorithm guarantees efficient mining performance by scanning the dataset only once, preventing overheads of pattern extraction based on the pruning strategies, and adopting fuzzy weight conditions to enhance the dependability of mining results. The extensive experimental results obtained from nine benchmark datasets indicate that our algorithm has outstanding performance in comparison to PADS and FPMax* algorithms.