A comparative study of an evolvability indicator and a predictor of expected performance for genetic programming

An open question within Genetic Programming (GP) is how to characterize problem difficulty. The goal is to develop predictive tools that estimate how difficult a problem is for GP to solve. Here we consider two groups of methods. We call the first group Evolvability Indicators (EI), measures that capture how amendable the fitness landscape is to a GP search. Examples of EIs are Fitness Distance Correlation (FDC) and Negative Slope Coefficient (NSC). The second group are Predictors of Expected Performance (PEP), models that take as input a set of descriptive attributes of a problem and predict the expected performance of GP. This paper compares an EI, the NSC, and a PEP model for a GP classifier. Results suggest that the EI does not correlate with the performance of the GP classifiers. Conversely, the PEP models show a high correlation with GP performance.