A pattern-aware method for maximal fuzzy supplement frequent pattern mining

Abstract

Advanced pattern mining to extract the hidden but useful information by using proper structure is vital important for efficient information mining in large-scale practical datasets. The existing algorithms have not been capable of effective solving the fuzziness uncertainty of items and confirming the appropriate structure of studied patterns. In order to generate more proper practical patterns, a base-(second-order-effect) pattern structure is proposed to represent the internal relationships among items. In addition, fuzzy weight constraints and properties have been presented to reflect the importance of uncertainty for each item in a whole dataset and in one transaction. Thus, the proposed maximal FSFPs mining algorithm guarantees efficient mining performance based on the proposed advanced pattern-aware dynamic search strategy, 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 six benchmark datasets indicate that our algorithm has outstanding performance in comparison to PADS and FPMax? algorithms.