Near-infrared prediction of edible oil frying times based on Bayesian Ridge Regression

Abstract

In the detection and analysis of edible oil quality, Total Polar Compound (TPC) has been widely used as an evaluation index of frying oil quality. Its content has been shown to have a linear relationship with the number of frying times. The more frying times, the higher the TPC content in the oil, the worse the quality of the oil. In this study, a Bayesian Ridge Regression (BRR) model is established to detect the frying times of soybean oil, peanut oil, and rapeseed oil based on near-infrared spectroscopy (NIRS) detection techniques. The study collects a total of 450 spectral signals from soybean oil, peanut oil and rapeseed oil. Then partial least regression (PLS), Support Vector Regression (SVR) and BRR prediction models are established to predict frying times of the three frying oils and furthermore to reflect the quality of frying oils. The results show that under the full spectra, the SVR and BBR prediction regression models established by the original spectra and the standard normalized variable (SNV) processed spectra can predict the frying times better, while the PLSR model cannot effectively predict the frying times in all cases. After selecting the characteristic wavelength through UFS, the BRR prediction model works best. The coefficient of determination (R2) and the root mean square error of prediction (RMSEP) of the model are 0.9852, 0.5463, respectively. In summary, the method of this study can effectively predict the number of frying times and provide an effective means for the rapid detection of frying oil quality.