Model Selection for PCA-Linear SVM for automated detection of NS1 molecule from Raman spectra of salivary mixture

Of recent, detection of Non-structural Protein 1 (NS1) in saliva has become appealing, as it may lead to a noninvasive detection method for NS1-related diseases at the febrile phase, before complication developed. NS1 is found to have a molecular fingerprint with the use of SERS technique. Our work here intends to determine an optimum PCA-Linear SVM model for automated detection of NS1 molecules from Raman spectra of NS1 adulterated saliva. Raman spectra of normal saliva (n=64) and saliva adulterated with low concentration NS1 (n=64) are used. Since Raman features extracted for each spectrum numbered at 1801, ranking and selection of features in order of their contribution is important prior to classification, for efficient computation. Hence, PCA for feature selection and SVM with linear kernel for classification are integrated. It is found that the Cattel´s Scree test is the best stopping criteria for PCA with a selection of 5 PCs and a box constraint of 20 is optimum for Linear SVM. Together they achieve a classification performance, [accuracy sensitivity, specificity], of [98.71% 98.97% 98.44%].