Explainable Pattern Learning in Exploring Robust Characteristics in Metaheuristic Design
Résumé
The Vehicle Routing Problem (VRP) is a complex optimization problem due to its NP-Hard nature, and it is mostly solved using metaheuristic algorithms. Recent developments in machine learning have demonstrated the potential to improve these approaches by substituting human-crafted designs with data-driven methods. Building on this advance, we examine the role of different characteristics or features in predicting the quality of VRP solutions, identifying several features that consistently serve as strong predictors and could be leveraged in the design of metaheuristic algorithms. We suggest that while feature importance can vary, specific characteristics or features remain reliable predictors across different scenarios.