Hybrid Sorting-Out Algorithm COMBI-GA with Evolutionary Growth of Model Complexity

Abstract

Paper presents latest achievements in the development of the sorting-out hybrid COMBI-GA algorithm based on a new mechanism of evolutionary growth of models complexity to find the optimal model structure. The mechanism uses generation of model structures of GA initial population by binomial random number generator with low probability and specific mutation operator with adding some units in model structures. The effectiveness of this mechanism is compared with two another mechanisms of evolutionary simplification (using binomial random number generator with big probability to form the GA initial population and mutation operator with reducing amount of units in model structures) and conventional random changing the complexity (using binomial random number generator with average probability of forming the GA initial population and mutation operator where gene values in a chromosome are inverted according to a given probability). The presented experimental results demonstrate that this new algorithm with evolutionary complication of model structures performs quickly, accurately and reliably when solving both artificial and real inductive modelling tasks.