One of the crucial issues in genetic programming is how to efficiently preserve promising building blocks; knowing that a common problem with genetic operators in tree- based genetic programming is that fit sub-trees can be subject to destruction by an inappropriate choice of cross-over or mutation point since the cross-over and mutation points selection are done at random. In order to minimize the destructive effects of the random selection of the sub-tree from the parent tree, and focus the action of mutation on the genomes that are not fully evolved, a new kind of mutation operators which attempt to remove a sub-tree with the least effect on the parent tree, is introduced. In this paper, we examine new forms of selective mutation that can be used in tree-based genetic programming. This is achieved through a comparative study, being done according to some performance indicators like the distance measure between the parents and their offspring. The efficiency of the operators is assessed by experimental analysis of genetic programs for symbolic regression problem.
展开▼