Guiding function set selection in genetic programming based on fitness landscape analysis

This paper attempts to provide a guideline for function set selection based on fitness landscape analysis. We used two well-known techniques, autocorrelation function and information content, to analysize the fitness landscape of each function set. We tested these methods on a large number of real-valued symbolic regression problems and the experimental results showed that there is a strong relationship between autocorrelation function value and the performance of a function set. Therefore, autocorrelation function can be used as a good indicator for selecting an appropriate function set for a problem.