An Ensemble Learning Approach for Predicting Sustainable Fuel Blend Properties

The prediction of fuel blend properties is a critical aspect of the petroleum and renewable energy industries for ensuring product quality, meeting regulatory standards, and optimizing production processes. Traditional methods rely on expensive laboratory testing and time-consuming experimental procedures. This paper proposes a stacking ensemble learning approach for accurately predicting multiple fuel blend properties simultaneously using advanced gradient boosting techniques. We implement a multi-output stacking regressor combining LightGBM, CatBoost, and Multi-Layer Perceptron (MLP) as base learners, with Ridge regression as the meta-learner. Our comprehensive evaluation using 5-fold cross-validation demonstrates that the ensemble approach achieves excellent predictive performance with a Mean Absolute Percentage Error (MAPE) of 1.47% across ten blend properties. The model successfully captures complex non-linear relationships between component compositions and resulting blend characteristics. The results validate the effectiveness of modern gradient boosting ensemble methods for fuel property prediction, offering practical implications for sustainable fuel formulation and real-time quality control in industrial settings.