Optimized Association Rule Mining with genetic algorithms

The mechanism for unearthing hidden facts in large datasets and drawing inferences on how a subset of items influences the presence of another subset is known as Association Rule Mining (ARM). There is a wide variety of rule interestingness metrics that can be applied in ARM. Due to the wide range of rule quality metrics it is hard to determine which are the most `interesting´ or `optimal´ rules in the dataset. In this paper we propose a multi-objective approach to generating optimal association rules using two new rule quality metrics: syntactic superiority and transactional superiority. These two metrics ensure that dominated but interesting rules are returned to not eliminated from the resulting set of rules. Experimental results show that when we modify the dominance relations new interesting rules emerge implying that when dominance is solely determined through the raw objective values there is a high chance of eliminating interesting rules.