Advancing Biodiesel Quality Assessment Using Machine Learning Algorithms: A Mini-Review on Oxidative Stability Prediction and Models Benchmarking

Abstract

This review compiled experimental data on antioxidant-stabilized biodiesel and subsequently conducted a comparative analysis of seven advanced machine learning (ML) algorithms, including AdaBoost, CatBoost, Voting Ensemble, XGBoost, LightGBM, Random Forest, and Neural Networks, to identify the optimal framework. A structured review methodology was employed to extract relevant data from studies conducted between 2004 and 2024, and thereafter, boosted the experimental data sets using SMOGN-based data augmentation methodology. The tested models were evaluated using standardized training/testing splits and appraisal metrics, including coefficient of determination (R2), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE). SHapley Additive exPlanations (SHAP) analysis was conducted for interpretability. The comparative analysis showed that AdaBoost achieved the best predictive accuracy and generalization with a test R2 of 0.928 and an RMSE of 2.42, outperforming other models. CatBoost and Voting Ensemble also performed well; however, XGBoost and Random Forest suffered from overfitting. Based on the findings, this work affirms the reliability of the tested models as effective platforms for stability prediction, directly addressing the currently felt industrial need for an alternative to time-consuming standardized tests for biodiesel stability measurement. Finally, this mini-review delivers key contributions, including synthesizing the current experimental landscape on biodiesel stabilization, recognizing feedstock as a primary stabilizing factor, and putting forward a clear data-driven framework for ML applications in this domain as a reliable tool for quality assessment and optimization.