Abstract
This study evaluates the effectiveness of various regression models in predicting the physicochemical properties of biochar, essential for sustainable agriculture and environmental remediation. A comprehensive review of recent literature compared the predictive accuracies of linear, non-linear regression (NLR), quadratic, and multiple linear regression (MLR) models. Findings highlight that MLR models perform exceptionally well, with R² values exceeding 0.92, particularly in predicting complex interactions like cation exchange capacity (CEC) and electrical conductivity (EC). NLR models also demonstrated strong performance, achieving high median R² values, especially in predicting High Heating Value (HHV), with R² values up to 0.9802. Pyrolysis Temperature (PT) was identified as a frequent and significant predictor for properties such as EC and nitrogen content. However, properties like CEC and Specific Surface Area (SSA) presented challenges due to inconsistencies between high R² and higher Root Mean Square Error (RMSE) values, indicating underlying variability. Municipal Solid Waste (MSW) biochar was the most challenging to predict due to its heterogeneous composition. This study advocates for integrating MLR with non-linear techniques to develop hybrid models, enhancing predictive accuracy and practical usability, and optimizing biochar utilization in agriculture and environmental remediation.
