DOE-based parameter tuning for local branching algorithm

Abstract

Design of experiments (DOE) refers to a process of planning the experiments so that appropriate data that can be analysed by statistical methods will be collected, resulting in valid and objective conclusions. This paper presents a DOE-based approach for parameter tuning of local branching algorithm. Local branching is a metaheuristic technique utilising a general MIP solver to explore neighbourhoods. This solution strategy is exact in nature, although it is designed to improve heuristic behaviour of MIP solver at hand. The proposed approach has been applied to find shortest Hamiltonian path in travelling salesman problem (TSP). A Hamiltonian path is a path in an undirected graph, which visits each node exactly once, and returns to the starting node. To evaluate the algorithm, the standard problems with different sizes are used. The performance of the algorithm is analysed by the quality of solution and CPU time.