Machine learning modeling for thermochemical biohydrogen production from biomass

Abstract

This paper outlines the steps for applying machine learning (ML) models to predict biohydrogen yields from biomass using thermochemical treatments. Input features include elemental compositions and thermochemical process parameters, while outputs are biohydrogen yields reported in existing studies. Procedures and software for performing ML modeling on biohydrogen yield predictions are provided. Input features were analyzed using Random Forest (RF) and eXtreme Gradient Boosting (XGB) models, interpreted through SHapley Additive exPlanations (SHAP) and Partial Dependence Plot analyses. XGB demonstrated superior performance over RF in predicting hydrogen yields, as measured by mean squared error values. Fixed carbon content, moisture, and volatile matter content significantly influenced the yields. Process temperature and fixed carbon content showed an increase in yield when temperatures were below 600 °C and carbon content was below 20%. The provided programs are adaptable for ML modeling and help users efficiently organize datasets to develop their models.