Pruning generalized rules for stock markets accumulated by Genetic Network Programming with Rule Accumulation

A new strategy on pruning rules accumulated by Genetic Network Programming with Rule Accumulation (GNP RA) has been proposed in this paper. The generalized rules extracted by training GNP are pruned by GA in the validation phase. Each rule has two variables: U and N. Variable U determines if the rule is used or not, while variable N shows that the information on N days is used. By mutating variables U and N of each rule, the portfolio of U and N is changed, as a result, the rules are pruned. The performance of the pruned rules is tested in the testing phase, meanwhile, the best mutation rates for variable U and variable N are also studied. The simulation results show that the pruned rules work better than the rules without pruning.