The authors develop a machine learning approach that predicts which edges are relevant for high-quality solutions to routing problems. This allows the search space of classical optimization methods to be systematically reduced, accelerating both mathematical solvers and metaheuristics. Results for the Traveling Salesman Problem and the Capacitated Vehicle Routing Problem show that up to approximately 90% of candidate edges can be removed while maintaining high-quality solutions.
New Publication in Networks: Using Machine Learning to Accelerate Optimization Methods
