Mining association rules with rare and frequent items. Issue 4 (1st January 2013)

• Record Type: Journal Article
• Authors: Sousa, Ricardo; Rodrigues, Fatima
• Is Part Of: International journal of knowledge engineering and data mining. Volume 2:Issue 4(2013)
• Page Start: 237
• Page End: 247
• Publication Date: 2013-01-01
• Subjects: data mining -- association rules -- frequent items -- rare items; Knowledge representation (Information theory) -- Periodicals; Data mining -- Periodicals; 006.305
• Languages: English
• ISSNs: 1755-2087

Abstract

Many existing association rule algorithms are based on the support-based pruning strategy to prune the combinatorial search space. This strategy is not effective for discovering interesting patterns, because low values of support generates too many rules, involving items with different support levels and poorly correlated, and high levels of support generates very few rules and generally trivial ones. In this paper we describe an algorithm to mining association rules with both rare and frequent items – MIRF algorithm. This algorithm does not require the minimum support to be specified in advance. Rather, it generates in each iteration all possible item sets, and extracts only those positively correlated in order to obtain a rule set whose size is smaller, easier to interpret and with both frequent and rare items. Also only the most relevant rule is extracted from each item set, which significantly reduces the time required for the mining process and the number of rules generated. Experimental evaluation of our algorithm on several databases will be presented.