Predicting algal biochar yield using eXtreme Gradient Boosting (XGB) algorithm of machine learning methods

Abstract Pyrolysis is a thermochemical pathway widely used for the conversion of biomass into useful products such as biochar, bio-oil, and syngases. A recent surge in the adoption of the pyrolysis process at realtime scenarios for the appropriate management and conversion of residues demands the modeling of the pyrolysis process. Prediction of algal biochar yield along with its composition was attempted in this study with the eXtreme Gradient Boosting (XGB) machine learning method. An extensive grid search method has been implemented in the XGB model to explore all the possible considered input parameter combinations for predicting the biochar yield. Thirteen different pyrolytically important input parameter combinations have been attempted and compared with the combination suggested by the feature selection technique of model for predicting the biochar yield. This feature selection technique highlights the H/C, N/C, ash content, pyrolysis temperature, and time as the key parameters on deciding the algal biochar yield, where H, C, N are hydrogen, carbon and nitrogen content of biomass. The highest regression coefficient (R2) of 0.84 has been achieved between experimental and model predictive biochar yield for the testing dataset, once the model was trained with the training dataset. Pearson correlation coefficient matrix unraveled the correlation among and in between input parameters and biochar yield. Feature Importance Plots revealed temperature as the most influential factor. SHapley Additive exPlanations (SHAP) Dependence Plots depicted the interactive effect of temperature and other input parameters on the algal biochar yield. Summary Plots showed the combined features of importance through feature and SHAP values. The developed XGB model provides new insights on comprehending the influence of input parameters on predicting the algal biochar yield.