Reducing Space Search in Combinatorial Optimization Using Machine Learning Tools

A new metaheuristic, called Feature-Guided MNS (FG-MNS) is proposed, combining well-known local search with simple machine learning techniques. In this metaheuristic, a solution is represented by features (mean depth of each route, standard deviation of the length of each route, etc.). The solver uses decision trees to define promising areas in the features space. The search is mainly focused on the promising areas, in order to minimize the exploration time, and to improve the quality of the found solutions. Additional neighborhoods, guided by the features are proposed.