Fitness Distance Correlation, as statistical measure of Genetic Algorithm difficulty, revisited

This paper revisits past works on fitness distance correlation (FDC) in relation to genetic algorithms (GA) performance, and puts forth evidence that this statistical measure is relevant to predict the performance of a GA. We propose an interpretation of Hamming-distance based FDC, which takes into account the GA dynamics and the effects of crossover operator. We base this proposition on the notion of predicates developed by M. D. Vose. The aim of this article is double since, using results obtained with the FDC, it confirms Vose's theory, and starting from this theory, it corroborates FDC as a relevant predictor measure of problem difficulty for GAs.