The Role of Representation on the Multidimensional Knapsack Problem by means of Fitness Landscape Analysis

Five encodings for the Multidimensional Knapsack Problem are investigated, using fitness landscape analysis techniques, in order to better understand the influence of genetic representations when solving a combinatorial optimization problem. Fitness distance correlation and autocorrelation measures are employed to analyze the encodings. The effect of heuristics, as well as repair and local optimization is also examined. The investigation helps to understand how the adopted representations influence the search performance of an evolutionary algorithm.