Development and Analytical Validation of Robust Near-Infrared Multivariate Calibration Models for the Quality Inspection Control of Mozzarella Cheese

Abstract

This paper proposed and validated robust diffuse reflectance near-infrared methods for the direct determination of fat and moisture in cow mozzarella cheeses using partial least squares regression. They were developed under the realistic conditions of routine analysis in a state laboratory of quality inspection control and were used for analyzing a great variety of mozzarella samples manufactured by different manufacturing procedures and originating from the whole state of Minas Gerais, Brazil (more than 100 different producers). A robust methodology was implemented, including the detection of outliers and the harmonization of the multivariate concepts with the traditional univariate guidelines. The models were constructed in the ranges from 38.7 to 58.0 % w/w on dry basis for fat and from 41.5 to 55.1 % w/w for moisture, providing root mean square errors of prediction of 2.1 and 0.9 %, respectively. Both methods were validated through the estimation of figures of merit, such as linearity, trueness, precision, analytical sensitivity, ruggedness, bias, and residual prediction deviation. Once the methods were adopted, their performances were monitored for approximately 1 year through control charts and were considered satisfactorily stable with prediction errors within the established limits. Beyond these specific methods, it was also pursued to present a complete methodology for multivariate analytical validation, an important aspect for the implementation of near-infrared spectroscopy methods in the routine of food quality inspection.