Risk Prediction in Design of Distributed Supply Chains Using Machine-Learning: A Bi-Objective Optimal Approach

In today's interconnected and volatile global economy, distributed manufacturing has become not only feasible but imperative. Traditional facility location models, typically static and deterministic, fail to capture the dynamic risks posed by geopolitical instability and socioeconomic volatility. To address this, a bi-objective Integer Linear Programming (ILP) model is developed that simultaneously minimizes cost and risk in supply chain facility decisions. The model permits flexible reallocation of existing capacities and explores new potential sites under various shift-based capacity expansion patterns. A core contribution of this work is the integration of a robust Machine Learning (ML) framework to estimate location-specific risk based on political, economic, and social factors. Using an extensive dataset covering 161 countries from 2006 to 2019, we construct a multiclass risk prediction pipeline that incorporates coordinates, political, economic, and social indicators. A comprehensive model selection exercise, spanning 23 feature-set configurations and 42 model hyperparameter combinations, is performed using a stratified ten-fold cross-validation approach. The Random Forest (RF) ensemble emerges as the top-performing classifier, followed closely by XGBoost (XGB), significantly outperforming neural networks in both accuracy and robustness. The RF model achieves a ten-fold cross-validated accuracy of over 96% and a macro F1-score of over 95%, indicating strong performance even on minority classes. These risk predictions are then embedded into the ILP formulation using a lexicographic preemptive optimization approach to resolve the trade-off between cost and risk, given the predictions made by RF for the years 2020-2024. A case study from the North American automotive sector is used to validate the proposed methodology. Results demonstrate that incorporating ML-based fragility forecasts into facility location planning enables risk-aware, cost-effective decision-making via an integrated decision-support framework for organizations navigating uncertain global supply chains.