Constructing and validating a novel hybrid ensemble regression framework to estimate biohydrogen production from dark fermentation

Biohydrogen production by dark fermentation offers a sustainable path for renewable energy and waste valorization. This study systematically combines the gradient boosted decision tree, kernel ridge regression, and soft gating, followed by residual correction to construct a novel framework to estimate the achievable biohydrogen yield by dark fermentation. This framework uses different operating conditions, including pH, hydraulic retention time, temperature, stirring rate, metabolite distribution, substrate type and concentration, and trace metals to estimate the achievable biohydrogen yield through dark fermentation. The proposed framework estimates 255 biohydrogen yields with an overall root mean square error of 0.0281 and correlation coefficients of higher than 0.997. Shapley-based analysis indicates that trace metal (Ni), substrate concentrations, Fe dose, and metabolite signature (butyrate) have the strongest influence on the model. The results approve the effectiveness and interpretability of the constructed role-aware hybrid framework for precisely simulating the behavior of biohydrogen production by dark fermentation. • A novel ML framework is developed for interpretable prediction of biohydrogen yield. • Model predicts biohydrogen yield from diverse substrates and operating parameters. • Model identifies substrate and trace metal as primary drivers of biohydrogen yield. • Model accurately predicts 255 actual data of biohydrogen yield, R > 0.997, MAE<0.0292. • Generalization and accuracy of model are better than existing tools in literature.