A novel genetic programming based classifier design using a new constructive crossover operator with a local search technique

Abstract

A common problem in genetic programming search algorithms is the destructive nature of the crossover operator in which the offspring of good parents generally has worse performance than the parents. Designing constructive crossover operators and integrating some local search techniques into the breeding process have been suggested as solutions. In this paper, we proposed the integration of variants of local search techniques in the breeding process, done by allowing parents to produce many off springs and applying a selection procedure to choose high performing off springs. Our approach has removed the randomness of crossover operator. To demonstrate our approach, we designed a Multiclass classifier and tested it on various benchmark datasets. Our method has shown the tremendous improvement over the other state of the art methods. © 2013 Springer-Verlag.