Efficient Interesting Association Rule Mining Based on Causal Criterion Using Feature Selection

Association rule mining often produces massive rules, many of which do not meet users’ interest. To address the problem of evaluating and sorting useful association rules, a variety of interestingness measures has been proposed during the past two decades. Nevertheless, current interestingness measures are inadequate to reveal the essential characters hidden in the data. Causality indicates not only that the variables are related, but also how varying a variable is likely to induce a change of another, therefore it is more useful in prediction and reasoning. In this paper, we introduce causality to measure the interestingness of association rules and give a formal definition of casual criterion. Furthermore, we propose a new framework for generating causal association rules using feature selection, and implement an algorithm to efficiently mine the causal association rules. Experiment result on both real and synthetic datasets show that the algorithm performs better than traditional algorithms, and can discover causal association rules from databases efficiently.