Exploring machine learning approaches for biohydrogen production through dark fermentation in wastewater
Ibrahim Shomope, Amaal Abdulraqeb Ali, Muhammad Tawalbeh, Amani Al–Othman, Fares Almomani
Abstract
The global dependence on fossil fuels continues to contribute to greenhouse gas emissions, driving the search for cleaner energy alternatives like biohydrogen. Dark fermentation has emerged as a promising method for sustainable hydrogen production while simultaneously treating wastewater. However, optimizing biohydrogen yields remains challenging due to the complexity of biological interactions and environmental factors. Machine learning (ML) offers a data-driven approach to predict and enhance hydrogen production efficiency. In this review, recent studies employing ML techniques are systematically analyzed to evaluate their role in modeling and optimizing biohydrogen generation through dark fermentation. This review examines various ML models, including artificial neural networks, support vector machines, decision trees, and gradient boosting techniques, for their effectiveness in optimizing fermentation conditions. Unlike traditional models like Monod kinetics, the anaerobic digestion model no.1 (ADM1), and response surface methodology (RSM), which are limited by fixed input ranges, results indicate that ML models outperform traditional statistical methods, with CatBoost achieving an R 2 of 0.98 and SVM reaching 0.988. Key influencing factors include chemical oxygen demand, nickel concentration, and butyrate levels. Furthermore, the review also highlights methodological gaps, prioritization of lifecycle assessments and cost-benefit analyses, and also provides insights into the future integration of ML with experimental workflows. While ML-driven optimization has significantly improved hydrogen yields, further research is required to refine models, expand datasets, and improve scalability for industrial applications.