A mining frequent itemsets procedure using a classification network

Association rules mining, one of the important data mining tasks, aims at uncovering all frequent itemsets among transactions. The frequent itemset mining algorithms such as Apriori, DHP and FP-growth suffer from the bottleneck of itemsets generation or the large number of times of database scans required for computing itemset support. In this paper, we propose a classification network, which can classify transactions in the form of binary vector, for efficiently locating maximum frequent itemsets (MFIs). The algorithm requires no candidate generation and needs only one round of database scan. An illustrative example and experiments were performed to demonstrate the efficiency of the proposed method.