Machine learning models for predicting calorific value of biomass from proximate and ultimate analyses across raw, briquette, and charcoal forms

Calorific value is among the most critical factors for assessing the quality of biofuels. Accurate estimation of biomass calorific value is essential for evaluating its suitability as a renewable energy source; however, conventional experimental determination is labor-intensive, costly, and time-consuming. This study developed machine learning (ML)-based predictive frameworks to estimate the calorific value of biomass using proximate and ultimate analyses across raw biomass, briquettes, and charcoal forms. A dataset comprising 600 biomass samples derived from 100 agricultural residues was compiled from published literature, incorporating raw, densified, and carbonized biomass materials. Four independent predictive frameworks were established using ultimate and proximate analyses. Multiple regression-based ML algorithms were comparatively evaluated. Model performance was assessed using root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²). Among the evaluated algorithms, KNN and SVR achieved the best predictive performance, with R² values of 0.883 and 0.991 for ultimate-analysis-based briquette and charcoal prediction, respectively, and 0.879 and 0.997 for proximate-analysis-based prediction, with consistently low prediction errors. Comparative assessment with previously published empirical equations demonstrated competitive predictive capability of the developed models. The proximate-analysis-based framework significantly exhibited superior predictive performance and stability, achieving an accuracy of 96.5%, compared with 97% for the ultimate-analysis-based framework. Overall, the proposed machine learning frameworks provide a rapid, interpretable, and cost-effective approach for biomass calorific value estimation, supporting efficient biomass screening and sustainable waste-to-energy applications. These findings provide a reliable framework for calorific value prediction, supporting biofuel production and sustainable agricultural waste management.