Comparison Between Linear and Nonlinear Machine-Learning Algorithms for Predicting the Properties of Biodiesel Using Near-infrared Spectra

This study points out the application of near-infrared (NIR) spectra combined with machine-learning approaches to evaluate biodiesel properties. The performance comparison between partial least squares regression (PLSR)-based linear and support vector regression (SVR)-based nonlinear machine-learning algorithms for predicting the biodiesel properties is the main objective of this paper. The models were built for four biodiesel properties: pH, viscosity, density, and water content. As a result, the PLSR had better performance than the SVR. An effective model of each biodiesel property prediction exhibited the coefficient of determination for the prediction (r2) and root mean square of prediction (RMSEP) of 0.89 and 0.01 mg KOH.g
^-1
, 0.75 and 0.07 cSt, 0.84 and 2.77 kg.m
^-3
, and 0.75 and 79.33 mg.kg
^-1
for pH, viscosity, density, and water content, respectively.