An evolutionary algorithm for cost-sensitive decision rule learning

Abstract

Most of classification learning methods aim at the reduction of the number of errors. However, in many real-life applications it is misclassification cost, which should be minimized. In the paper we propose a new method for cost-sensitive learning of decision rules from datasets. Our approach consists in modifying the existing system EDRLMD (Evolutionary Decision Rule Learner with Multivariate Discretization). EDRL-MD learns decision rules using an evolutionary algorithm (EA). We propose a new fitness function, which allows the algorithm to minimize misclassification cost rather than the number of classification errors. The remaining components of EA i.e., the representation of solutions and the genetic search operators are not changed. The performance of our method is compared to that of C5.0 learning system. The results show, that the modified EDRL-MD is able to effectively process datasets with non-equal error costs.