Towards Scalable Algorithm for Closed Itemset Mining in High-Dimensional Data

Abstract

Mining frequent itemsets from large dataset has a major drawback in which the explosive number of itemsets requires additional mining process which might filter the interesting ones. Therefore, as the solution, the concept of closed frequent itemset was introduced that is lossless and condensed representation of all the frequent itemsets and their corresponding supports.  Unfortunately, many algorithms are not memory-efficient since it requires the storage of closed itemsets in main memory for duplication checks. This paper presents BFF, a scalable algorithm for discovering closed frequent itemsets from high-dimensional data. Unlike many well-known algorithms, BFF traverses the search tree in breadth-first manner resulted to a minimum use of memory and less running time. The tests conducted on a number of microarray datasets show that the performance of this algorithm improved significantly as the support threshold decreases which is crucial in generating more interesting rules.