Design and Validation of a Predictive Agribusiness Intelligence Platform for U.S. Food-Supply and Market Resilience

Abstract

This quantitative study designed and validated a predictive agribusiness intelligence platform for assessing U.S. food-supply and agricultural-market resilience through the integration of agricultural production, inventory, weather, logistics, international trade, commodity-market, and macroeconomic indicators. A longitudinal predictive-analytics design was applied to 18,720 commodity-region-month observations representing 26 major agricultural commodities across 48 contiguous U.S. states from January 2015 through December 2024. The final dataset retained 95.1% of initially integrated observations and achieved 97.8% overall data completeness. Statistical analysis incorporated descriptive statistics, correlation analysis, analysis of variance, multivariable regression, effect-size estimation, and risk modeling, while predictive performance was evaluated using Random Forest, Support Vector Machine, Gradient Boosting, XGBoost, and neural-network architectures. Mean Food-Supply Resilience was 72.8 ± 11.6, while Market Resilience averaged 69.5 ± 13.2. XGBoost produced the strongest independent-test performance for Food-Supply Resilience, achieving an R² of 0.914, RMSE of 4.38, classification accuracy of 92.8%, F1-score of 91.9%, and AUC of 0.967. The neural network performed most strongly for Market Resilience, achieving an R² of 0.889, RMSE of 5.06, accuracy of 91.4%, F1-score of 90.9%, and AUC of 0.958. Production and yield contributed 24.8% of normalized predictive importance, followed by inventories and storage at 21.6%, weather and climate at 18.7%, transportation and logistics at 15.3%, international trade at 10.8%, and input and macroeconomic conditions at 8.8%. Low inventory coverage increased the adjusted odds of high or critical food-supply risk by 3.18 times, while extreme commodity-price volatility increased market-resilience risk by 3.46 times. Concurrent production, inventory, and logistics stress produced the largest combined food-supply effect (OR = 5.82). Independent validation, repeated cross-validation, calibration assessment, subgroup analysis, and sensitivity testing demonstrated comparatively small generalization gaps and stable predictive performance. The findings validated the platform as a multidimensional quantitative approach for identifying, classifying, and predicting variation in U.S. food-supply and agricultural-market resilience across commodities, regions, and disruption conditions.