Depth-First Uncertain Frequent Itemsets Mining based on Ensembled Conditional Item-Wise Supports

Uncertain frequent pattern mining is usually challenged by the single probabilistic frequent threshold or the single expected support as the measurements of frequent itemsets. A promising solution based on multiple expected minimum support has been introduced in more recent studies to distinguish the mining values of each item, but the intrinsic combinatorial explosion still limited this strategy to be further improved for more generic scenarios. In this paper, a novel mining scheme for uncertain frequent itemsets is proposed. By ensembling multiple conditional item-wise supports, the problems of information redundancy as well as loss caused by a single probabilistic frequent threshold can be effectively improved. Furthermore, by using a variety of pruning strategies based on the property of sorted downward closure and the concept of least minimum probabilistic frequent threshold, an UFP-ECIS (Uncertain Frequent Pattern Mining with Ensembled Conditional Item-wise Supports) algorithm is also introduced. Substantial experiments have been proved to demonstrate that the proposed mining scheme and algorithm has enhanced the information precision of the uncertain frequent itemsets mining.