Is it worth to approximate fitness by machine learning?

It is usual to need an approximate model in evolutionary computation when fitness function is deemed to be abstract or expected to have a long computation time. In these cases, research on possibility of fitness approximation should proceed before applying an evolutionary algorithm in real-world problems. In this paper, it was found that we could train machine learning algorithms with the sampled solutions when problem size is large, if there is a possibility of fitness approximation at small problem sizes.