Optimizing methane catalytic hydrogen production via a hybrid Boruta-XGB and stacking ensemble machine learning framework

Hydrogen production from methane is a crucial technology for the transition to clean energy, but conventional catalyst development relies on costly and time-consuming trial-and-error experiments. The objective of this study is to enhance the methane catalytic hydrogen production process by employing machine learning methodologies to augment hydrogen yield and mitigate experimental expenses. The machine learning model was constructed by the collection of 1772 data points from the extant literature. The improved Boruta algorithm was employed for the purpose of feature screening, and the prediction model was constructed by combining the Stacking integrated learning method. This method is capable of revealing the effects of process control parameters and catalyst design on the methane hydrogen production process by means of systematic analysis of the relationship between the input parameters and the output parameters. SHAP and PDP interpretation tools were then used to reveal the effects of process parameters and catalyst design on hydrogen production and to identify key influential features, such as Al₂O₃ content, pore size (PS), surface area (SA) and Time. The findings demonstrate that the machine learning models developed are capable of precise prediction of hydrogen production, with the Stacking model exhibiting superior prediction performance in the test set, as evidenced by an R² value of 0.9544, an RMSE value of 5.8601, and an MAE value of 3.0731. This study provides an efficient tool for the screening and optimization of methane-hydrogen-producing catalysts, with industrial applications being a key focus. In addition to offering a practical guide for industrial applications, the study provides theoretical support that helps to elucidate the mechanism of methane hydrogen production, thereby promoting the development of more efficient catalysts.