An analysis of training models to evolve heuristics for the travelling salesman problem

Abstract

Designing heuristics is an arduous task, usually approached with hyper-heuristic methods such as genetic programming (GP). In this setting, the goal of GP is to evolve new heuristics that generalise well, i.e., that work well on a large number of problems. To achieve this, GP must use a good training model to evolve new heuristics and also to evaluate their generalisation ability. For this reason, dozens of training models have been used in the literature. However, there is a lack of comparison between different models to determine their effectiveness, which makes it difficult to choose the right one. Therefore, in this paper, we compare different training models and evaluate their effectiveness. We consider the well-known Travelling Salesman Problem (TSP) as a case study to analyse the performance of different training models and gain insights about training models. Moreover, this research opens new directions for the future application of hyper-heuristics.