Comparing Ensemble and Standalone Machine Learning Models for Optimizing Biodiesel Production via Esterification.

The increasing energy demand and environmental problems caused by fossil fuel consumption have increased the demand for sustainable and green energy sources. Biofuels stand out as alternative fuels that are renewable, biodegradable, and environmentally friendly. However, traditional experimental methods used in biofuel production are restrictive in both time and cost in determining and optimizing process parameters. Therefore, modeling approaches based on machine learning (ML) techniques have recently stood out to improve production efficiency. In this study, the biodiesel production process based on esterification reaction of oleic acid was modeled and optimized by using four distinct modelsMultilayer Perceptron (MLP), AdaBoost SVR, Gradient Boosting, and a high-performance Voting Regressor (an ensemble of the three). The model utilizes key process parameters including reaction time (1-6 h), catalyst loading (3-18% by weight), and a methanol/oleic acid molar ratio (3:1-15:1) to predict oleic acid conversion (%). Model performances were evaluated using R 2, mean absolute percentage error (MAPE), and mean square error (MSE) values obtained from training, testing, and cross-validation data sets. Among the models, the Voting Regressor achieved the highest test R 2 value (0.9653), with relatively low test MAPE (2.27%) and MSE (4.7557) values, as well as the highest cross-validation R 2 value (0.9513). Parametric analysis using the Voting Regressor model showed that increasing catalyst loading and methanol/oleic acid molar ratio improved conversion up to a certain level, after which a plateau tendency was observed. Under optimized conditions (5.49 h, 18 wt % catalyst, and 15:1 molar ratio), a maximum conversion of 99.23% was predicted. The results demonstrate that ensemble learning can significantly improve the reliability of biodiesel process modeling while reducing experimental workload in chemical engineering applications.