Explainable Machine Learning for Risk-Driven Forecasting of Global Food Commodity Prices: Geopolitical, Logistic, and Energy Drivers

Escalating geopolitical tensions, supply chain disruptions, and energy market volatility have intensified global food price instability, posing serious risks to food security. To address this challenge, this study proposes an explainable forecasting framework that integrates historical price dynamics with external drivers, applied to five essential food commodities, olive oil, eggs, chicken, beef, and bread, over the period 1990-2023. The analysis examines how logistics efficiency, energy costs, and geopolitical uncertainty relate to food prices through complex, nonlinear relationships. Going beyond conventional models, the proposed nonlinear autoregressive neural network with exogenous inputs (NARX) captures these dynamics while preserving explainability: through an ablation-style comparison against a baseline autoregressive model, it systematically evaluates alternative panels of exogenous inputs to identify those with the greatest forecasting contribution. The results show that the predictive relevance of exogenous panels varies across commodities and evaluation metrics, and underscore the value of transparent, multidimensional analysis for improving food price forecasting and understanding price dynamics in an increasingly uncertain global context.