Efficient Mining of Maximal Frequent Itemsets Based on M-Step Lookahead

We propose a new maximal frequent itemset mining algorithm, named m-step lookahead. This is a variant of the Max-Miner algorithm that, instead of counting the support of the largest possible superset of each candidate itemset, counts the support of a superset with a predetermined length. This is designed to circumvent the weakness in the Max-Miner algorithm that the probability of finding a frequent superset is extremely low for the first several passes. By looking for a smaller superset, m-step lookahead may find long frequent patterns more quickly than Max-Miner. Our experimental results demonstrate that this is the case for certain datasets and user-defined parameters.