An improved model for estimating optimal VRP solution values

Since it is computationally expensive to solve the vehicle routing problem (VRP) optimally, as this problem is NP-hard, in this technical note we study how to accurately approximate the optimal VRP tour length. In our previous papers, we developed a linear regression model including the mean and standard deviation of the modified Clarke and Wright heuristic solution values, which was able to predict the optimal VRP tour length fairly well. In this note, we find that by doing a small amount of extra work to include the minimum of the modified Clarke and Wright heuristic solution values, we can improve the predictive results substantially.