FTIR spectroscopy coupled with machine learning approaches as a rapid tool for identification and quantification of artificial sweeteners

Abstract

Fourier transform infrared (FTIR) spectroscopy calibrations were developed to simultaneously determine the multianalytes of five artificial sweeteners, including sodium cyclamate, sucralose, sodium saccharin, acesulfame-K and aspartame. By combining the pretreatment of the spectrum and principal component analysis, 131 feature wavenumbers were extracted from the full spectral range for modelling to qualitative and quantitative analysis. Compared to random forest, k nearest neighbour and linear discriminant analysis, support vector machine model had better predictivity, indicating the most effective identification performance. Furthermore, multivariate calibration models based on partial least squares regression were constructed for quantifying any combinations of the five artificial sweeteners, and validated by prediction data sets. As shown by the good agreement between the proposed method and the reference HPLC for the determination of the sweeteners in beverage samples, a promising and rapid tool based on FTIR spectroscopy, coupled with chemometrics, has been performed to identify and objectively quantify artificial sweeteners.

Keywords: Artificial sweeteners; Chemometrics; Identification; Mid-infrared spectroscopy; Quantification.