An Analysis of Semantic Aware Crossover

It is well-known that the crossover operator plays a very important role in genetic programming (GP). It is also widely admitted that standard crossover is made mostly randomly without semantic information. The lack of semantic information is the main reason that causes destructive effect, generally producing children worse than parents, of standard crossover. Recently, we have proposed a new semantic based crossover for GP, that is called Semantic Aware Crossover (SAC) [26]. It was shown in [26] that SAC outperforms standard crossover (SC) in solving a class of real-value symbolic regression problems. This paper extends [26] by giving some deeper analyses to understand why SAC helps to improve the performance of GP in solving these problems. The analyses show that SAC can increase the semantic diversity of population and this helps to reduce the crossover destructive effect in GP. The results also show that although SAC requires more time for checking semantics, this extra time is negligible