Differential evolution-optimized hybrid ensemble machine learning for biogas production from food waste: Implications for agricultural circular bioenergy systems

Anaerobic digestion of food and agricultural waste provides a promising pathway for renewable biogas production and integration of organic residues into circular bioenergy systems. However, accurate prediction of biogas production remains challenging because of the complex and nonlinear interactions among operational and biological variables. This study presents a differential evolution-optimized hybrid ensemble machine learning framework for predicting biogas yield from food waste anaerobic digestion processes. A comprehensive dataset containing six key process variables, including hydraulic retention time, pH, temperature, volatile solids content, organic loading rate, and reactor volume, was used for model development. Four machine learning approaches (least squares support vector machine, Gaussian process regression, multilayer perceptron, and polynomial regression) were employed as base learners, while a general regression neural network served as the meta learner in an ensemble architecture. Differential evolution was used to optimize model hyperparameters and improve predictive performance. Model interpretability was investigated using SHAP analysis, which identified pH, hydraulic retention time, and reactor volume as the most influential variables. The proposed ensemble model consistently outperformed all individual models. It achieved R values of 0.99255 and 0.98491, NSE values of 0.98512 and 0.96865, WIA values of 0.99622 and 0.99170, RMSE values of 0.023 and 0.033, and MAPE values of 4.73 and 6.90 for the cross-validation and testing datasets, respectively. Results demonstrate the potential of the proposed framework for accurate and reliable biogas yield prediction of anaerobic digestion systems. Improved prediction of biogas production can support the efficient conversion of food and agro-food organic residues into renewable energy.