PCA Kernel Based Extreme Learning Machine Model for Detection of NS1 from Salivary SERS Spectra.

Use of NS1 as a biomarker in saliva has led to the non-invasive and early detection of Flaviviridae related diseases. Saliva is preferred as medium of detection because of its advantages such as non-invasive, painless and easy to collect. Work here intends to compare the performance of KELM classifier with linear and RBF kernels for classification of NS1 from salivary SERS spectra. Prior to KELM, PCA with different termination criteria (Cattle Scree test, CPV and EOC) are used to extract important features and reduce the dimension of SERS spectra dataset. Regularization coefficient (C-value) for linear kernel and Regularization coefficient (C-value) and \( \gamma \)-value for RBF kernel are varied to find the optimum KELM classifier model. For linear kernel, 100% accuracy, precision, sensitivity, specificity is achieved for Linear model with EOC criterion and C-value set to 0.1, 0.2, 0.5, 1 and 2. For RBF kernel, 100% performance of accuracy, precision, sensitivity and specificity is achieved with RBF model with EOC criterion and values of 0.04, 0.06, 0.08, 0.1, 0.2, 0.4, 0.6, 0.8 and 1. The C-value is fixed to 1. The best Kappa value of 1 is obtained when all performance indicators scored 100%. For both Linear-KELM and RBF-KELM, EOC termination criterion gives the highest performance. It also observed that KELM classifier is data dependent.