Machine Learning-Based Modeling of Catalytic Anaerobic Digestion for Biogas Yield Prediction

Catalytic enhancement of anaerobic digestion (AD) has emerged as a promising strategy to improve biogas production efficiency; however, the complex, nonlinear interactions between catalyst type, loading, and process variables make reliable yield prediction a significant challenge. This study develops and compares multiple machine learning (ML) predictive models including Artificial Neural Networks (ANN), Adaptive Neuro-Fuzzy Inference Systems (ANFIS), and Gradient Boosting to predict biogas yield in catalytically enhanced AD systems. Input variables include catalyst type (Ni-based, zeolite, biochar), catalyst loading concentration, pH, temperature, organic loading rate (OLR), hydraulic retention time (HRT), and substrate composition, with training and validation datasets compiled from peer-reviewed literature retrieved from Scopus and Web of Science. Model performance is evaluated using R², RMSE, and MAE metrics, and SHAP (Shapley Additive Explanations) analysis is applied to identify the most influential process variables. Preliminary findings indicate that ANN and ANFIS models achieve prediction accuracy above R² = 0.93 when catalyst-related features are included as inputs, with SHAP analysis revealing that catalyst loading and HRT are among the top contributors to model output variance. This computational study demonstrates that ML models trained on literature-derived datasets can effectively capture the influence of catalytic variables on biogas yield, providing a scalable, resource-efficient tool for process optimization and guiding future experimental design in catalytic AD research.