Distributed approximate mining of frequent patterns

This paper discusses a novel communication effcient distributed algorithm for approximate mining of frequent patterns from transactional databases. The proposed algorithm consists in the distributed exact computation of locally frequent itemsets and an effective method for inferring the local support of locally unfrequent itemsets. The combination of the two strategies gives a good approximation of the set of the globally frequent patterns and their supports. Several tests on publicly available datasets were conducted, aimed at evaluating the similarity between the exact result set and the approximate ones returned by our distributed algorithm as well as the scalability of the proposed method.