| چکیده انگلیسی مقاله |
Milk is wholesome nutritious dairy product, which has been confronted with various adulterations. Hydrogen peroxide (H2O2) is widely used for disinfection purposes by food industry enterprises, and it is one of the major adulterants in milk and dairy industry. Even though due to chemical processes inside raw milk, it can contain small quantities of H2O2 (1–2 mgL-1), the concentration must be 10 times higher in order to destroy pathogens. At the same time, high concentrations of H2O2 in milk can lead to changes in its chemical composition which, in turn, can lead to negative effect for the consumers [1]. Therefore, the main objective of this work is to use a handheld near infrared spectrometer (NIRS) as a rapid and non-destructive technique in combination with ensemble learners to detect H2O2 adulterant in milk. For this purpose, authentic raw bovine milk samples were collected from two dairy farms located in Iran. After pooling the samples, appropriate amount of a standard solution of H2O2 were added at 10 concentration levels ranged 0.05% to 18% (v/v). A handheld NIRS device (Tellspec Inc., Toronto, Canada) was used to acquire NIR spectra in the reflectance mode in the spectral range of 900-1700 nm. First of all, the pattern of the milk samples (pure and adulterated) and of the 256 NIR variables was explored by principal component analysis (PCA). The first two principal components explained 98% of the total variance. Then, the dataset was split into training (70%) and test (30%) sets using duplex algorithm. Additionally, a five-fold randomized cross-validation was used for internal model validation. Random subspace discriminant ensemble (RSDE) [2, 3] was used for classification. The performance of the RSDE method was evaluated in terms of sensitivity (Sen), specificity (Spe), and accuracy (Acc). All the values for Sen, Spe and Acc were above 95% in each case, which showed the reliability and robustness of the developed model. Furthermore, the RSDE method outperformed partial least squares-discriminant analysis (PLS-DA) and support vector machine (SVM) as common classification techniques used in food authenticity. |
