Ensemble machine learning as a mathematical approach to predict high-voltage electric field-assisted removal of color and oxidative indices from soybean oil using the activated porous electrodes

Abstract

Porous electrode materials provide a large surface area for electrochemical reactions and facilitate ion transport efficiently. By optimizing the design and properties of porous electrodes, it is possible to enhance the bleaching performance. In this research, three machine learning (ML) algorithms, namely Support Vector Regression, Ridge Regression, and Decision Tree Regression were utilized for predicting the bleaching efficiency under different conditions of HVEF-assisted bleaching (HVEFAB). Subsequently, the outcomes of these models were compared to ensemble models created by combining the algorithms as mentioned earlier using three techniques including bagging, boosting, and stacking. The levels of chlorophyll and carotenoids were reduced as a result of increased voltage, modification in the electrode surface, number of cathode-anode, and the presence of an inert electrolyte. Meanwhile, the reduction of red color and chlorophyll pigments was more significant than yellow color and carotenoid pigments. Moreover, the application of HVEFAB can potentially decrease the levels of polar compounds, such as oxidative compounds on the electrode surface. Based on the MSE, MAE, and R2 metrics, the algorithms' efficacy in forecasting outputs 1 (color index) and 2 (oxidative index) can be ordered as follows: Stacking > Boosting > Bagging > Decision Tree (DT) > Ridge > Support Vector Regression (SVR). The findings demonstrate that integrating these three models, mainly through the stacking technique, substantially improves the accuracy of the predicting model. Particle swarm optimization (PSO) and polynomial regression rather than RSM represented a superior model to predict red color removal.