On the Use of Ant Programming for Mining Rare Association Rules

Most researches in association rule mining have focused on the extraction of frequent and reliable associations. However, there is an increasing interest in finding reliable rules that rarely appear, and recently, some classical solutions have been adapted to this field. The problem is that most of these algorithms follow an exhaustive approach, which have the drawback of becoming unfeasible when dealing with high complex data sets. This kind of problem can be also addressed as an optimization problem, for which bio-inspired algorithms have proved their ability. To this end, this paper presents an ant-based automatic programming method for discovering rare association rules. This algorithm lacks the drawbacks of exhaustive approaches, having also some advantages, such as the employment of a context-free grammar that allows to adapt the algorithm to a particular domain. Results show that this proposal can mine a set of reliable infrequent rules in a short period of time.