Integrating machine learning-based edge classification into ant colony optimization for the travelling salesman problem

Machine learning is increasingly acknowledged as an essential component across multiple domains, including optimization. This article presents an enhanced Ant Colony Optimization (ACO) algorithm that integrates machine learning to solve the Travelling Salesman Problem (TSP). Building upon two previously developed frameworks – Adaptive Ant Colony Optimization with Node Clustering and an ML-based edge classification model – we propose a novel method, ACO-ML, that embeds ML predictions directly into the optimization process. Specifically, the ML model is used to: (1) initialize pheromone values based on edge probabilities, (2) guide the clustering of nodes, and (3) influence transition probabilities during solution construction. The main contribution in artificial intelligence lies in the principled integration of predictive models within a metaheuristic algorithm, resulting in a data-informed and adaptive optimization framework. A comprehensive ablation study demonstrates how machine learning enhances the decision-making logic of ACO at multiple stages, improving convergence behavior and solution quality while maintaining computational efficiency. From the engineering perspective, the proposed method addresses real-world routing challenges, such as Unmanned Aerial Vehicle (UAV) mission planning and military logistics optimization, where fast and high-quality solutions are critical. The algorithm is evaluated on diverse TSPLIB benchmark instances and consistently outperforms both the original algorithm and state-of-the-art ACO variants. In addition, a comparative analysis with recent pure learning-based TSP solvers indicates that the proposed hybrid approach remains highly competitive while preserving the robustness of a metaheuristic search framework. These findings confirm that embedding predictive models within metaheuristic structures can significantly improve practical optimization performance in mission-critical environments.