Features Selection and Rule Removal for Frequent Association Rule Based Classification

The performance of association rule based classification is notably deteriorated with the existence of irrelevant and redundant features and complex attributes. Association rules naturally often suffer from a large volume of rules generated, many of which are not interesting and useful. Thus, selecting relevant feature and/or removing unrelated rules can significantly improve the association rule performance. In this paper, we explored and compared feature selection measures to filter out irrelevant and redundant features prior to association rules generation. Rules that encompassed with irrelevant/redundant features were removed. Based on the experimental results, removing rules that hold irrelevant features slightly improve the accuracy rate and capable to retain the rule coverage rate.