Machine learning–based methane yield prediction using a structured anaerobic digestion dataset

Accurate prediction of methane yield is essential for optimizing anaerobic digestion (AD) systems and improving the efficiency of agricultural biomass-to-energy conversion. This study presents a machine learning–based predictive framework trained on a structured and experimentally derived dataset encompassing physicochemical feedstock properties, operational parameters, and biogas performance indicators. The dataset includes more than 500 labeled samples representing major agricultural residues, characterized by Total Solids (TS), Volatile Solids (VS), C/N ratio, lignocellulosic composition, temperature, pH, and Organic Loading Rate. Six supervised learning algorithms like Gradient Boosting Regressor (GBR), Light GBM, Cat Boost, Extra Trees, K-Nearest Neighbors (KNN), and Elastic Net were developed and evaluated using an 80/20 train–test split, five-fold cross-validation, and performance metrics including RMSE, MAE, and R². Results indicate that Light GBM achieved the highest predictive accuracy with an R² of 0.95 and the lowest RMSE, demonstrating the dataset’s strong feature representation and model suitability. Feature importance analysis revealed Volatile Solids, lignin content, and C/N ratio as the most influential predictors of methane yield. The findings confirm that machine learning models, when trained on well-structured AD datasets, can significantly enhance methane yield estimation and support intelligent, data-driven biogas plant optimization. This study establishes a scalable framework for predictive AD modeling and offers a foundation for integrating AI-driven decision-making into sustainable waste-to-energy systems.