An Empirical Analysis on Renewable Energy: Biogas Production Prediction Using Machine Learning

Biogas is gaining prominence as a renewable energy source with significant potential to reduce greenhouse gas emissions and mitigate environmental impacts associated with fossil fuels. This study presents an improved biogas production estimation method using machine learning (Ridge Regression, Lasso Regression, Random Forest, XGBoost, LightGBM, and GBM) combined with explainable AI (XAI) techniques to enhance model interpretability. Our rigorous evaluation using Wilcoxon Signed-Rank Tests demonstrated that LightGBM and XGBoost consistently outperformed other algorithms—LightGBM achieved superior performance in the 70:30 train-test split (RMSE = 0.075, R2 = 0.895), while XGBoost excelled in both 80:20 (RMSE = 0.091, R2 = 0.847) and 50:50 splits. These models proved significantly better than traditional methods (p < 0.05) in all comparisons, with particularly strong performance against linear regression approaches (p-values as low as 0.001). The analysis identified waste efficiency and total daily waste (kg/day) as the most critical predictive features. Despite dataset limitations (U.S.-only livestock data, n = 344), these findings offer valuable guidance for biogas professionals and investors to optimize production forecasting and operational management strategies for improved renewable energy outputs. The research demonstrates how machine learning can enhance both prediction accuracy and interpretability in renewable energy applications.