Detection of adulterated milk using two-dimensional correlation spectroscopy combined with multi-way partial least squares

• We study 2D correlation spectra (2DCOS) of melamine-tainted milk and pure milk.
• Multi-way PLS discriminant analysis model(NPLS-DA) is built based on 2DCOS of milk.
• NPLS-DA method can effectively classify the adulterated milk and pure milk.
• N-PLS quantitative model for melamine is built based on 2DCOS of adulterated milk.
• N-PLS model has stronger predictive capability than traditional PLS model.

A new approach for qualitative and quantitative analysis of adulterated milk is proposed by combining two-dimensional correlation spectroscopy (2DCOS) with multi-way partial least squares (N-PLS). Forty pure milk and 40 adulterated milk samples with melamine were prepared and the infrared absorption spectra were measured. The two-dimensional (2D) correlation spectra were calculated to construct the multi-way partial least squares discriminant analysis (NPLS-DA) model for milk adulterated with melamine. Our study showed the higher accuracy of predicting adulterated milk of 95% using NPLS-DA, versus 85% using conventional one-dimensional spectra and partial least squares discriminant analysis (PLS-DA). Also, the quantitative analysis models were constructed to determine concentration of melamine in milk. The root mean square error of prediction (RMSEP) was 0.106 g L−1 using the N-PLS method and 0.16 g L−1 using conventional one-dimensional spectra and partial least squares (PLS). Comparison results show that the proposed new method is superior to the traditional method and the use of 2DCOS with N-PLS is promising to quantify the adulterants in pure milk.