Discovering high utility itemsets with multiple minimum supports

• Published in: Intelligent data analysis Vol. 18; no. 6; pp. 1027 - 1047
• Main Authors: Ryang, Heungmo, Yun, Unil, Ryu, Keun Ho
• Format: Journal Article
• Language: English
• Published: London, England SAGE Publications 01.01.2014
• Subjects: Algorithms; Construction; Data mining; Data processing; Diseases; Medical; Trees; Utilities; Frequent itemsets; multiple minimum supports; rare frequent itemsets; utility mining
• ISSN: 1088-467X; 1571-4128
• DOI: 10.3233/IDA-140683

Generally, association rule mining uses only a single minimum support threshold for the whole database. This model implicitly assumes that all items in the database have the same nature. In real applications, however, each item can have different nature such as medical datasets which contain information of both diseases and symptoms or status related to the diseases. Therefore, association rule mining needs to consider multiple minimum supports. Association rule mining with multiple minimum supports discovers all item rules by reflecting their characteristics. Although this model can identify meaningful association rules including rare item rules, not only the importance of items such as fatality rate of diseases but also attribute of items such as duration of symptoms are not considered since it treats each item with equal importance and represents the occurrences of items in transactions as binary values. In this paper, we propose a novel tree structure, called MHU-Tree (Multiple item supports with High Utility Tree), which is constructed with a single scan. Moreover, we propose an algorithm, named MHU-Growth (Multiple item supports with High Utility Growth), for mining high utility itemsets with multiple minimum supports. Experimental results show that MHU-Growth outperforms the previous algorithm on both real and synthetic datasets, and can discover useful rules from a medical dataset.