RPP Algorithm: A Method for Discovering Interesting Rare Itemsets

Sadeq Darrab, David Broneske, Gunter Saake

Abstract: The importance of rare itemset mining stems from its ability to discover unseen knowledge from datasets in real-life domains, such as identifying network failures, or suspicious behavior. There are significant efforts proposed to extract rare itemsets. The RP-growth algorithm outperforms previous methods proposed for generating rare itemsets. However, the performance of the RP-growth degrades on sparse datasets, and it is costly in terms of time and memory consumption. Hence, in this paper, we propose the RPP algorithm to extract rare itemsets. The advantage of the RPP algorithm is that it avoids time for generating useless candidate itemsets by omitting conditional trees as RP-growth does. Furthermore, our RPP algorithm uses a novel data structure, RN-list, for creating rare itemsets. To evaluate the performance of the proposed method, we conduct extensive experiments on sparse and dense datasets. The results show that the RPP algorithm is around an order of magnitude better than the RP-growth algorithm.