Near Infrared Spectroscopy Combined with Chemometrics and Supervised Machine Learning for Detection of Water Adulteration in Raw Bovine Milk

Abstract

Spectroscopic methods and chemometrics/machine learning offer a fast, easy, and simple approach to detect adulteration in milk. The aim of this study was to evaluate the performance of several commonly used classification and regression algorithms for detecting milk adulteration, using added water as the adulterant, based on near-infrared (NIR) spectroscopy. Different water/milk mixtures containing 1% to 40% added water were prepared and analyzed. Supervised classification algorithms such as linear discriminant analysis (LDA), partial least squares discriminant analysis (PLS-DA), and support vector machines discriminant analysis (SVM-DA) were applied. Subsequently, supervised regression algorithms for quantitative detection, such as partial least squares regression (PLSR) and support vector regression (SVR), were evaluated. The PLS-DA models using autoscaling (AS) and standard normal variate (SNV) showed the best performance for validation in terms of sensitivity (SEN), specificity (SPE), accuracy (ACC), and error with values of 0.953, 0.850, 0.920, 0.098, and 0.930, 0.950, 0.937, 0.060, respectively. For quantitative determination, the best performance was achieved using partial least squares regression (PLSR) combined with standard normal variate (SNV) preprocessing of the spectral data. This model exhibited a validation root mean square error (RMSECV) of 3.912, a prediction Root Mean Square Error (RMSEP) of 2.117, a residual predictive deviation (RPD) of 6.15, and a coefficient of determination for prediction (R²P) of 0.972. These results demonstrate that the combination of NIR spectroscopy with chemometric and machine learning techniques, such as PLSR, is suitable for the quantification of added water in raw milk samples.