Comparative Performance Analysis of Machine Learning Models for Prediction of Rapeseed Oil Methyl Ester Biodiesel Production

Abstract

To forecast the biodiesel production from rapeseed oil methyl ester (RSOME), this work makes use of several machine learning (ML) techniques, using Gaussian process regression (GPR), gradient boosting with DT (GBDT) and multilayer perceptron (MLP). To optimise the RSOME manufacturing process, we improved the models' output while keeping their generalisability and increasing the accuracy of their predictions. The variables that were selected as inputs for the biodiesel production process include the Reaction temperature - A (degC), catalyst amount - B (% wt.), treatment duration - C (min.) and methanol to rapeseed oil molar ratio - D, all of which impact the transesterification. At the phase boundary, NaOH was utilised as a catalyst for the response of rapeseed oil with quick-chain alcohols. When compared to the mean absolute percentage error (MAPE) criterion, the obtained error rate for MLP, GBDT, and GPR model is 0.09211, 0.21536 and 0.0842 respectively. The achieved R2 value for MLP, GBDT and GPR process is 0.973, 0.991 and 0.997 respectively. With a Mean Absolute Error (MAE) of 4.7, the GPR is determined to be the optimal model. The proposed method yielded an ideal RSOME production value of around 99.97% for A (64degC), B (0.8 wt. %), C (8 min) and D (12).