Predicting the methane production of microwave-pretreated anaerobic digestion of food waste: A machine learning approach

Anaerobic digestion (AD) is a widely adopted waste management strategy that transforms organic waste into biogas, addressing both energy and environmental challenges. Feedstock pretreatment is crucial for enhancing organic matter breakdown and improving biogas yield. Among various techniques, microwave (MW) irradiation-based pretreatment has shown significant promise. However, the optimization of MW-assisted AD processes remains underexplored, necessitating predictive tools for process simulation. Machine Learning (ML) has recently emerged as a powerful alternative for predicting and optimizing AD performance. In this study, an ML-driven pipeline was developed to predict methane yield based on food waste (FW) composition, AD reactor parameters, and MW pretreatment conditions. A range of data preprocessing techniques and ML models (linear, non-linear, and ensemble) were systematically evaluated, with model performance assessed via hyperparameter-optimized cross-validation. The most accurate models (non-linear and ensemble) achieved R2 > 0.91 and RMSE <35 mL/g volatile solids (gVS), whereas linear models underperformed (R2 < 0.71, RMSE >70 mL/gVS). Support Vector Machine (SVM) emerged as the best-performing model, with R2 ∼0.94 and RMSE ∼34 mL/gVS. Beyond predictive accuracy, this study offers novel insights into MW pretreatment's role in AD efficiency. Permutation feature importance (PFI) analysis revealed that while MW pretreatment enhances methane yield, its effects are secondary to reactor pH and FW composition. This suggests that MW treatment primarily facilitates substrate disintegration but does not drastically alter biochemical methane potential unless coupled with optimized reactor conditions. Additionally, minor fluctuations in MW pretreatment time and temperature were found to have negligible impacts on methane production, indicating a level of operational flexibility in MW-based AD processes. These findings provide a refined understanding of MW pretreatment's practical implications, guiding process design for improved scalability and industrial application.