Predicting biochar yield from biomass pyrolysis: A comprehensive data-driven approach using machine learning and SHAP analysis
Walid Abdelfattah, Munthar Kadhim Abosaoda, Krunal Vaghela, J Gowrishankar, Prabhat Kumar Sahu, Kamred Udham Singh, R. Sivaranjani, Rohit Chauhan, Siya Singla, Samim Sherzod
Abstract
Biochar, a pivotal product of biomass pyrolysis, is crucial in sustainable agriculture, carbon sequestration, and renewable energy systems. This study employs a comprehensive data-driven approach to predict biochar yield using machine learning algorithms. The dataset, comprising 14 chemical, physical, and reaction parameters collected from reputable studies, was processed using outlier detection via the Monte Carlo Outlier Detection (MCOD) algorithm and hyperparameter tuning to optimize model performance. Among the evaluated models, the Decision Tree algorithm emerged as the most robust predictor, achieving an R² value of 0.771, a mean square error (MSE) of 16.182, and an average absolute relative error percentage (AARE%) of 11.848% in testing. The SHAP analysis identified residence time, pyrolysis temperature, and ash content as the most influential predictors of biochar yield. Simplified explanations and interpretability offered by these models support their practical application. These findings underline the potential of machine learning to optimize biochar production, enhance agricultural systems, and create scalable methods for sustainable waste management, directly contributing to renewable energy and environmental remediation industries.