Precise prognostics of biochar yield from various biomass sources by Bayesian approach with supervised machine learning and ensemble methods

Abstract

ABSTRACT Biomass pyrolysis is a sustainable process for generating biochar from agricultural waste, though it is generally energy-intensive and time-consuming. To address this issue, the researchers gathered data from published literature on various biomass types and employed ensemble methods (LSBoost) and supervised machine learning (Gaussian process regression) to construct predictive models. The results reveal that both models can predict well, with excellent correlations between expected and actual values. In comparison to the LSBoost model (0.9783 for training and 0.9879 for testing), the Gaussian process regression (GPR) model had higher R values for training (0.9883) and testing (0.9969). Likewise, the R2 values during training (0.9767) and testing (0.9938) were greater in the case of the GPR model than for the LSBoost model (0.9571 for training). Nash-Sutcliffe efficiency (NSE) revealed that both models captured the data precisely. However, the GPR model outperformed the LSBoost model in both during training as well as model test stages, providing higher (0.9766 for training and 0.9933 for testing) values. The GPR model outperforms the others due to superior correlation, improved variability capture, and lower errors. These findings offer useful insights for sustainable biomass utilization and provide valuable insights for optimizing pyrolysis operations.