- Research Article
- 10.1016/j.indcrop.2026.122778
Interpretable machine learning algorithms for calculating the heat conduction coefficient of crop-derived renewable fuels
- Mar 01, 2026
- Industrial Crops and Products
- Walid Abdelfattah + 7 more +7
Accurate prediction of the heat conduction coefficient (HCC) of renewable fuels is essential for improving thermal management and advancing clean combustion technologies. Existing empirical correlations lack robustness across wide thermodynamic and molecular domains, creating a significant research gap. This study applies six machine‑learning techniques, Support Vector Machine (SVM), Artificial Neural Networks (ANN), Decision Tree (DT), Bagged Tree (BT), Least Squares Boosting Trees (LSBT), and Genetic Programming (GP), to develop generalized HCC models using 2051 high‑quality experimental data points described by reduced pressure, reduced temperature, and molecular weight. Among these models, SVM demonstrated the strongest predictive capability, achieving an overall Mean Absolute Percentage Error (MAPE) of 0.96 % and a coefficient of determination (R²) of 98.78 % on unseen data. Complementarily, GP provided a transparent analytical correlation with MAPE of 1.97 % and R² of 95.18 %, offering a novel interpretable alternative to black‑box models. Additional evaluations, cross‑plot analysis, violin–box deviation profiling, five-fold cross‑validation, and training‑size generalization tests, confirmed SVM’s robustness, stability, and minimal sensitivity to dataset partitioning. Trend analysis further revealed physically consistent dependencies of HCC on pressure, temperature, and molecular chain length, while the SHapley Additive exPlanations (SHAP) interpretability demonstrated that reduced temperature is the dominant factor, followed by molecular weight and reduced pressure. Overall, this research delivers both high‑accuracy predictive models and a practically deployable mathematical correlation, bridging engineering usability with modern machine learning performance. The findings offer a comprehensive and reliable framework for estimating HCC in renewable fuels across broad operating conditions. • Machine learning models developed for accurate prediction of biofuel HCC. • SVM achieved superior accuracy with MAPE below 1 % across wide conditions. • GP delivered an explicit, interpretable HCC correlation for engineering use. • SHAP analysis revealed temperature as the dominant factor governing HCC. • Robust validation confirms reliability across diverse esters and regimes.
Read more