Application of Partial Least Squares-Near Infrared Spectral Classification in Diabetic Identification

In order to identify the diabetic patients by using tongue near-infrared (NIR) spectrum, a spectral classification model of the NIR reflectivity of the tongue tip is proposed, based on the partial least square (PLS) method. 39sample data of tongue tip's NIR spectra are harvested from healthy people and diabetic patients, respectively. After pretreatment of the reflectivity, the spectral data are set as the independent variable matrix, and information of classification as the dependent variables matrix, Samples were divided into two groups, i.e. 53 samples as calibration set and 25 as prediction set, then the PLS is used to build the classification model The constructed modelfrom the 53 samples has the correlation of 0.9614 and the root mean square error of cross-validation (RMSECV) of 0.1387. The predictions for the 25 samples have the correlation of 0.9146 and the RMSECV of 0.2122. The experimental result shows that the PLS method can achieve good classification on features of healthy people and diabetic patients.