Comparative Analysis of Machine Learning Models for Predicting and Optimizing Biodiesel Production Yield: A Study of Neural Networks, Random Forest, and Decision Tree Algorithms

This study compares three machine learning algorithms (Multilayer Perceptron Neural Network (MLP), Random Forest (RF), and Decision Tree (DT)) for modeling biodiesel production. For this purpose the synthesis methods (UIMS, MS, FPUI, PUI), the methanol to oil ratio (۳:۱ to ۱۵:۱) and reaction times (۵–۵۰ minutes), were considered as input parameters and the percentage of biodiesel production was considered as the output of the model. According to the results, the MLP model demonstrated superior predictive performance, with an R² score of ۰.۹۸۰۰, RMSE of ۳.۲۸, and MAE of ۲.۳۵, significantly outperforming RF (R² = ۰.۸۸۹۲) and DT (R² = ۰.۸۵۰۰). Also, the neural network model represents that all parameters (reaction time, methanol to oil ratio, and synthesis method) hold nearly equal importance. Based on the neural network model, the optimal synthesis conditions are: the UIMS method, a reaction time of ۴۷ minutes, and a methanol-to-oil ratio of ۵.۸:۱, yielding a predicted conversion of ۹۸%.

Keywords:

Biodiesel Production , Machine Learning , Neural Networks , Random Forest , Decision Tree , Process Optimization