Associative classification based on Mutually Associated pattern

Recent studies have shown that associative classification is a promising classification method. However, when the minimum support is too low, associative classification often generates a large set of rules, which results in two main challenges: (1) how to select an appropriate subset of rules to build a classifier; and (2) how to select a best rule for classifying new instances. In this paper, we propose a new associative classification approach called ACMA (Associative Classification based on Mutually Associated pattern). It is distinguished from other associative classification algorithms in two aspects. First, in order to reduce the number of rules, ACMA selects mutually associated patterns to generate rules, and also exploits information entropy of items to reduce research space. Second, ACMA employs a new rule ranking method which considers mutual association between the itemset and the predictive class. Our experiments on six UCI data sets show that ACMA approach is an effective classification technique, and has better average classification accuracy in comparison with CBA.