Machine learning models for the prediction of total yield and specific surface area of biochar derived from agricultural biomass by pyrolysis

Organic biomass pyrolysis to produce biochar is a viable approach to sustainably convert agricultural residues. The yield and SSA of biochar are contingent upon the biomass type and pyrolysis conditions, and their quantification necessitates the investment of time, energy, and resources. Therefore, in this study, data from 46 different types of biomass were extracted from the published literature and modeled based on a supervised machine learning approach with five different regression algorithms to predict the total yield and SSA of biochar. In general, the collected data were processed using a data exploration technique to remove outliers. The correlation between input variables was examined using the Pearson correlation coefficient method to eliminate highly correlated input variables, and the assorted data was further imputed for developing predictive models. The yield and SSA of biochar were predicted by feature importance analysis to reduce the computational complexity and latency of the model. Out of the 14 input variables, 9 were selected based on feature importance and redundancy, wherein pyrolysis temperature demonstrated the greatest relative importance of 33.6% in predicting targets. Compared to other models developed to predict total biochar yield and SSA, Random Forests performs better, having a maximum R2 value of 85% and a minimum absolute root mean squared error (RMSE) for both biochar yield and SSA. Therefore, the developed models could help predict total biochar yield and SSA for a variety of agricultural biomasses without the need for complex and energy-intensive pyrolysis experiments.