Combining semantically-effective and geometric crossover operators for genetic programming

We propose a way to combine two distinct general patterns for designing semantic crossover operators for genetic programming: geometric semantic approach and semantically-effective approach. In the experimental part we show the synergistic effects of combining these two approaches, which we explain by a major fraction of crossover acts performed by geometric semantic crossover operators being semantically ineffective. The results of the combined approach show significant improvement of performance and high resistance to a premature convergence. © Springer International Publishing Switzerland 2014.