An empirical evaluation of three popular meta-heuristics for solving Travelling Salesman Problem

Abstract

Metaheuristic algorithms are applied in various fields to solve realistic problems. In many situations, a researcher moves in perplexed situation when it comes to selection of an appropriate metaheuristic algorithm for any specific problem. To overcome from such situation a comparative study is must. Considering this view we have done the performance evaluations of three popular metaheuristic algorithms: Evolution Strategy, Tabu Search and Variable Neighborhood Search. We framed three research questions to evaluate our hypothesis. Extensive experiments are conducted and results are collected. It was observed that Variable Neighborhood Search approach performed far better than other approaches. But this result seems insufficient in presenting some conclusion. Therefore, various statistical tests such as F-test, Post-hoc tests were performed. An obvious outcome of this study is that there is an interaction effect between the problem sizes and the metaheuristic used and no clear superiority of one metaheuristic over the other.