Preprocessing Fitness Functions

Genetic algorithms, Evolutionary strategies and many variants of local search algorithms are being applied directly on a fitness function. Such direct search algorithms require minimal knowledge of a problem and can be applied to a wide variety of different domains. This paper suggests to apply a simple mapping on the fitness function – which we refer to as a “preprocessing” step. Any direct search algorithm is then applied on the modified function. We demonstrate empirically that our method, when applied to the Maximum Satisfiability Problem (MAXSAT) and Travelling Salesman Problem (TSP), improves the performance of a (1+1) EA. keyword: Fitness Landscape, Preprocessing, MAXSAT, TSP, Selection