Selection of vehicle routing heuristic using neural networks

Due to its combinatorial structure, the vehicle routing problem has been attacked by many heuristic solution approaches. Given the large number of available heuristics, selecting the best heuristic for a particular problem poses it own kind of difficulty. This study uses a simple neural network approach to select the best heuristic for a VRP instance according to its basic characteristics. The approach has been trained and tested on a large test bed which covers problems with a wide variety of characteristics. It was also tested on a set of benchmark problems from the literature. For these problems, a simple procedure was used to extract the problem characteristics from the problem data. Statistical analysis reveals that the performance of each heuristic is affected differently by the problem characteristics. Neural network results for both test sets show that our approach is capable of selecting the best algorithm for a given VRP instance.