Investigating role of interestingness measures in rule mining by embedding in novel soft computing method

Association rule mining is a data mining method that extracts correlations, frequent patterns and causal dependencies between attributes of datasets. Association rules can be used for data classification as they provide a comprehensible and intuitive way to categorize data. The effectiveness of a rule is typically measured using support-confidence framework. This paper studies several other objective measures of the worthiness of a rule and tries to determine their suitability for classification of data. The impact of these different measures with respect to accuracy of rules and size of rules is analyzed by combining them with a novel soft computing based rule mining method. Section I introduces the task of association rule mining and the soft computing method. Section II provides a review of literature pertaining to rule mining techniques. Section III discusses computation of several interestingness measures. Section IV provides description of proposed algorithm and conducted experiment. Section V discusses results. Section VI concludes the paper with some directions for further work.