Game-theoretic SHAP-driven interpretable forecasting of air cargo demand using Bayesian-optimized random forests

Reliable forecasting of air cargo demand is crucial for optimizing logistics operations, scheduling air freight capacity, and reducing operational costs in a dynamic global supply chain environment. This study proposes a novel interpretable forecasting framework that integrates Bayesian-optimized Random Forests (BO-RF) with game-theoretic SHAP (SHapley Additive exPlanations) analysis to enhance both prediction accuracy and model transparency. The proposed BO-RF model leverages Bayesian Optimization to fine-tune hyperparameters efficiently, thus improving the generalization performance of Random Forests on small-sample air cargo datasets. To address the interpretability challenge of machine learning models, SHAP values are introduced, providing theoretically grounded, fair attribution of each input feature’s marginal contribution based on cooperative game theory. Experiments based on real-world monthly air cargo data demonstrate that the proposed method outperforms traditional machine learning benchmarks in both accuracy and interpretability. By combining Bayesian-optimized ensemble learning with SHAP-based interpretability, the study contributes to the growing literature on explainable, data-driven forecasting in transportation and provides actionable insights for demand management and capacity planning in the air freight industry.