Efficient heuristic for very large 0/1 multiple knapsack problems

Abstract

In this paper, we describe a heuristic that solves very large problem instances (one million items) of the 0/1 multiple knapsack problem. The proposed method is based on IRT heuristic and it is a pure local search technique that attempts to improve within a subset of items instead of the complete set of n items. Each time an interval of items is randomly selected to choose the set of items that can increase the total profit. The experimental work reveals an excellent scalability to solve very large problem instances against IRT and Martelo&Toth heuristic method (MTHM).