Intelligent Optimization of Supply Chain Disruption Prediction: A Particle Swarm Optimization-Enhanced Random Forest Framework with Explainable Analytics

The need for intelligent optimization of supply chain operations has become imperative for organizations seeking to adapt quickly to an increasingly complex logistics environment. The complexity of predicting delivery delays stems from the large number of interdependent operational and external factors that may affect delivery performance. In this study, we propose a hybrid machine learning framework that employs both the Random Forest Classifier (RF) and the Particle Swarm Optimization (PSO) to provide more accurate predictions of delivery delays through automated hyperparameter tuning. We evaluated our proposed PSO–RF model using a publicly available benchmark dataset of shipment records, along with their respective operational characteristics and contextual risk indicator values. In the comparative experiments PSO–RF produced the best single configuration among the metaheuristic-tuned models, reaching an accuracy of 94.6% and a Matthews correlation coefficient of 0.739 at a swarm size of 10. This advantage is specific to the small-swarm setting: at swarm sizes of 20 and 40 the four optimizers examined the (PSO, SCA, GWO, DE) cluster within a narrow MCC band of 0.71–0.74, and SCA–RF is the most consistent, so no optimizer is uniformly superior across swarm sizes. Against the conventionally configured baseline classifiers the improvement is far larger and holds in every configuration, which indicates that the principal gain arises from optimizing the Random Forest hyperparameters rather than from the choice of search algorithm. Additionally, to provide actionable supply chain intelligence, we employed SHapley Additive exPlanations (SHAP) to conduct model interpretability analyses. The SHAP analysis identified that the most influential features within the model’s delay classification decisions include shipping costs, order weights, geopolitical risks, scheduled lead-time days, transportation mode, and base lead-time days.