The k‐coloring fitness landscape

This paper deals with the fitness landscape analysis of the k‐coloring problem. We study several standard instances extracted from the second DIMACS benchmark. Statistical indicators are used to investigate both global and local structure of fitness landscapes. An approximative distance on the k‐coloring space is proposed to perform these statistical measures. Local search operator trajectories on various landscapes are then studied using the time series analysis. Results are used to better understand the behavior of metaheuristics based on local search when dealing with the graph coloring problem.