Integrated Approach between Machine Learning and Operations Research: A Case Study in Grain Logistics

Abstract

Operations research (OR) and machine learning (ML) techniques, when used in isolation, are often inadequate for addressing complex optimization problems. While OR struggles with combinatorial complexity and imprecise and high-dimensional data, ML fails to ensure strict constraint compliance. This work proposes a hybrid methodology that integrates OR and ML by embedding clustering mechanisms within an optimization framework. The method iteratively applies global, intercluster, and intracluster optimization models using dynamic parameters such as agitation and temperature to explore the solution space while satisfying clustering and feasibility criteria. A modified K-means adaptively selects clusters around elbow-based values, considering the size and performance constraints. Applied to a global agricultural production case study, the approach achieves a nearly 95% decrease in computation time without compromising solution quality. These findings suggest that the proposed integration is a promising direction for addressing complex constrained optimization problems.