Variable selection coupled to PLS2, ANN and SVM for simultaneous detection of multiple adulterants in milk using spectral data

Fourier transform infrared (FT-IR) spectroscopy combined with chemometric methods was used to detect multiple adulterants in milk samples simultaneously. PLS-DA (partial least squares discriminant analysis) and SVM (support vector machine) were used for the 100% accurate classification of samples to differentiate the adulterants. RCGA (real coded genetic algorithm) was used to obtain 20, 30, and 40 different fingerprint wavenumbers from milk FT-IR spectra when spiked with starch, urea, and sucrose. Amongst the four algorithms tested, the performance of LS-SVM was observed to be superior having higher values for correlation coefficient (R-p(2)) for prediction of 0.9843, 0.9763, and 0.9964 and lower root-mean-square error of prediction (RMSEP) of 0.4197, 0.2617, and 0.3771 for starch, urea, and sucrose, respectively. RCGA was established as an efficient feature selection algorithm for obtaining user-defined fingerprints. Also, LS-SVM was demonstrated as a robust non-linear regression algorithm for simultaneous detection of milk adulterants. (C) 2021 Elsevier Ltd. All rights reserved.