PREDICTIVE MODELING OF BIODIESEL YIELD AND FUEL VISCOSITY BASED ON COMPARATIVE MACHINE LEARNING

The increasing demand for sustainable and environmentally friendly fuels has intensified research into biodiesel production, with particular emphasis on optimizing process efficiency and ensuring fuel quality. Among the key performance indicators, biodiesel yield reflects production effectiveness, whereas fuel viscosity governs combustion behavior and compliance with international fuel standards. This study presents a comparative ML–based predictive modeling framework for accurately predicting biodiesel yield and fuel viscosity using experimental process data. Six machine learning algorithms RF, DT, XGBoost, Linear Regression, Ridge Regression, and ANN were evaluated to assess their predictive capability across both response variables. The results demonstrate that tree-based ensemble models consistently outperform the linear and neural network approaches. For biodiesel yield prediction, RF achieved the best overall performance with a correlation coefficient of approximately 0.64 and the lowest root mean square error (RMSE ≈ 12.5%), indicating moderate but superior predictive accuracy relative to other models. In contrast, viscosity prediction exhibited significantly higher accuracy, with XGBoost emerging as the optimal model, explaining over 80% of the variance (R² ≈ 0.80) and achieving the lowest prediction errors (RMSE ≈ 1.6 mm²/s, MAE ≈ 1.0 mm²/s). These findings highlight that viscosity is more directly governed by measurable process parameters than biodiesel yield, which is influenced by additional unmeasured factors. Overall, this study confirms the effectiveness of ensemble ML models for biodiesel process modeling, particularly under limited experimental datasets. Future research should expand the dataset size, incorporate additional influential features such as feedstock composition, and explore hybrid modeling frameworks that integrate mechanistic knowledge with data-driven approaches, such as advancements will further enhance predictive robustness and support intelligent, sustainable biodiesel production systems.