AI-Driven Forecasting of U.S. Biofuel Production and Feedstock Demand: A Machine Learning Analysis

Abstract

Biofuel production in the United States is shaped by shifting policy requirements, fluctuating commodity markets, and changes within agricultural supply systems. Forecasting production volumes and feedstock needs is important for decisions made by producers, refiners, and agricultural planners. This study applies machine learning to predict monthly ethanol and biodiesel output alongside demand for key feedstocks such as corn and soybean oil. The analysis integrates data from federal energy and agricultural sources with market signals, weather conditions, and policy indicators. Several models are evaluated, including gradient boosted decision trees, long short-term memory networks, and hybrid ensembles, and their performance is compared against standard econometric baselines using cross-validated error metrics. The results show that machine learning models capture nonlinear relationships that conventional approaches fail to represent, leading to lower errors across short and medium-term forecasting windows. Feature analyses indicate strong influence from feedstock prices, planted acreage, refinery utilization patterns, and Renewable Fuel Standard volumes. The forecasts point to continued growth in ethanol production and moderate gains in biodiesel output, accompanied by rising demand for corn and soybean oil. The overall findings show that data-driven forecasting can improve planning and risk assessment in an evolving bioenergy sector.