A genetic algorithm with sequential niching for discovering small-disjunct rules

This work addresses the well-known classification task of data mining. In this context, small disjuncts are classification rules covering a small number of examples. One approach for coping with small disjuncts, proposed in our previous work, consists of using a decision-tree/genetic algorithm method. The basic idea is that examples belonging to large disjuncts are classified by rules produced by a decision-tree algorithm (C4.5), while examples belonging to small disjuncts are classified by a genetic algorithm (GA) designed for discovering small-disjunct rules. In this paper we follow this basic idea, but we propose a new GA which consists of several major modifications to the original GA used for coping with small disjuncts. The performance of the new GA is extensively evaluated by comparing it with two versions of C4.5, across several data sets, and with several different sizes of small disjuncts.