Using Meta-learning to Classify Traveling Salesman Problems

In this paper, a meta-learning approach is proposed to suggest the best optimization technique (s) for instances of the Traveling Salesman Problem. The problem is represented by a dataset where each example is associated with one of the instances. Thus, each example contains characteristics of an instance and is labeled with the name of the technique (s) that obtained the best solution for this instance. Since the best solution can be obtained by more than one technique, an example may have more than one label. Therefore, the metal earning problem is addressed as a multi-label classification problem. Experiments with 535 instances of the problem were performed to evaluate the proposed approach, which produced promising results.