An improved version of the frequent itemset mining algorithm

This paper presents an improved version of the Frequent Itemset Mining algorithm. Along with its generalization, this algorithm for association rule discovery was designed to be used in parallel and distributed environments. The improvements made to the core formulas have a substantial impact on the overall performance of the algorithm, by reducing to a bare minimum the candidate generation across the entire chain of processing nodes, without missing any potential valid candidates. These modifications make an exclusive use of the computations already performed in previous steps by other nodes in the processing chain in order to avoid generating redundant or otherwise useless invalid candidates.