Risk Management in Design of Distributed Supply Chains Using a Bi-Objective Optimization and Machine-Learning Approach
Abstract
In this day and age, distributed manufacturing is a fact of life. While basic facility location models are typically static and deterministic, the environment in which facilities operate is quite volatile due to factors such as the various encountered risks, population dynamics, market trends, distribution costs, and demand patterns. As a result, in the design of modern supply chains, adopted strategies may need to be revised and redistribution of the supply chain relocating a company’s manufacturing facilities may need to take place accordingly. To address this challenge, a Mixed-Integer Linear Programming (MILP) employing a bi-objective cost/risk function is developed. A lexicographic preemptive optimization approach is followed to solve the model using the CPLEX commercial solver. To predict risk associated with a geographic location, Machine Learning (ML) is utilized. The dataset used in this study covers the period from 2006 to 2021 and represents 139 countries in an unbalanced panel. Several ML models are considered to predict the degree of fragility of a geographic location based on political, economic, and social factors. Random Forests have demonstrated superior performance with an accuracy of 96% and an F1-score of 95%. It is believed that this combined ML and optimization approach is deemed to be of benefit to decision makers in design and improvement of their supply chains.