First-level tabu search approach for solving the multiple-choice multidimensional knapsack problem

We apply tabu search to the multiple-choice multidimensional knapsack problem. An initial solution involves a simple greedy heuristic. A neighbourhood is generated and a move is selected based on the best feasible neighbouring solution. The search continues for a fixed number of iterations and tabu search structures are used to improve the search process. We study the problem structure of available, benchmark, test problem instances and develop a new, more diverse set of test problem instances whose results better generalise to practice than do results obtained using existing test problems. We report the results of testing our tabu search approach versus existing solution approaches as applied to the available and our new test problem instances. The computational results show that our proposed approach performs comparably or better than the legacy heuristic approaches and clearly demonstrates that test problems affects how well results generalise to practice.