[4Yin-B] Poster 7

[4Yin-B-56] Discovery of Itemset with Probability Distribution Table as Statistical Background through Individuality-based Fuzzification

〇Eisuke Ooka¹, Kaoru Shimada¹ (1. Gunma University)

Keywords:

individuality,itemset,knowledge discovery,evolutionary computation

ItemSB (Itemset with statistically distinctive background) has been studied as an extension of association rule mining and frequent itemset mining in data mining. A discretization method has been proposed that exploits the fact that the data are similar to the target individual. However, this conventional method considers only whether data are similar and does not take the degree of similarity into account. In this study, we propose a method that incorporates the degree of similarity into the discretization process. Furthermore, we extend ItemSB by integrating this similarity-aware discretization, enabling more fine-grained utilization of similarity information. In addition, we propose ICPD(Itemset with Class Probability Distribution Table as Statistical Background). Experiments showed that ICPD enables novel types of data analysis that have not been achievable with existing approaches. The proposed method is expected to be applicable to causal discovery and related tasks.