Improving Direct Counting for Frequent Itemset Mining

During the last ten years, many algorithms have been proposed to mine frequent itemsets. In order to fairly evaluate their behavior, the IEEE/ICDM Workshop on Frequent Itemset Mining Implementations (FIMI’03) has been recently organized. According to its analysis, kDCI++ is a state-of-the-art algorithm. However, it can be observed from the FIMI’03 experiments that its efficient behavior does not occur for low minimum supports, specially on sparse databases. Aiming at improving kDCI++ and making it even more competitive, we present the kDCI-3 algorithm. This proposal directly accesses candidates not only in the first iterations but specially in the third one, which represents, in general, the highest computational cost of kDCI++ for low minimum supports. Results have shown that kDCI-3 outperforms kDCI++ in the conducted experiments. When compared to other important algorithms, kDCI-3 enlarged the number of times kDCI++ presented the best behavior.