Making a state-of-the-art heuristic faster with data mining

Hybrid metaheuristics—developed based on the combination of metaheuristics with concepts and techniques from other research areas—represent an important subject in combinatorial optimization research. Data mining techniques have been coupled with metaheuristics in order to obtain patterns of suboptimal solutions, which are used to guide the search for better-cost solutions. In this paper, we incorporate a data mining procedure into a state-of-the-art heuristic for a specific problem in order to give evidences that, when a technique is able to reach an optimal solution, or a near-optimal solution with little chance of improvements, the mined patterns could be used to guide the search for the optimal or near optimal solution in less computational time. We developed a data mining hybrid version of a previously proposed and state-of-the-art multistart heuristic for the classical $$p$$ p -median problem. Computational experiments, conducted on a set of instances from the literature, showed that the new version of the heuristic was able to reach optimal and near-optimal solutions, on average, 27.32 % faster than the original strategy.