Research of Uninformative Variable Elimination method for spectral data analysis of milk

During the NIR spectral analysis to quickly determine concentrations of essential components of milk, spectral region is wider, peaks are overlap, and searching space is larger, spectrum acquiring often subjects to interference coming from environmental noise and interference of other components, so it is necessary to make optimum selecting to wavelength variables. In this paper, for concentration of protein in milk samples, Uninformative Variable Elimination (UVE) method is used to make optimum selecting to wavelength variables, and the selected wavelengths are taken as the input variables to build PLS model. Prediction results of PLS model built after processed by the UVE method is respectively compared with results of the PLS model by Genetic Algorithms (GA) method and results of the PLS model without making wavelength variable selection. The result shows that the UVE-PLS model has great advantage comparing with the GA-PLS model, and using the UVE method to select the wavelength variable of milk spectrum can make variable numbers for the final PLS model become smaller, redundant information become minimize, robustness of model become steady, and testing time of spectrum acquiring become shorter.