Hand-Foot-Mouth Disease Classification using Features from Fibre Grating Biosensor Spectral Data

Hand, Foot and Mouth disease (HFMD) is a common viral childhood disease affected by the family of enterovirus and Coxsackie. Current laboratory identification is based on the RT-PCR test, which is expensive, time-consuming, and unsuitable for the pandemic. The SPR- TFBG was biofunctionalized with monoclonal antibody (Mab). Mab is a bioreceptor with an affinity for the virus for detecting EV-A71. A dataset of reflectance spectra of 660 samples of different virus impurities measured with SPR-TFBG biosensor to detect EV-71 virus was developed. The extracted signal has around 4000 different features based on wavelength information. The first subset was selected based on the region of interest analysis, and the dimension has reduced from 4000 to 1496 features. The dimensionality of the large feature set is reduced based on the statistical feature engineering procedure using 10 features including mean, variance, skewness, RMS, kurtosis, standard deviation, range, crest factor, impulse factor and shape factor. Subsequently, classification of the virus (signal) data is achieved through SVM and it is evaluated with different types of kernels. For the evaluation of classifiers, we used accuracy, sensitivity, precision and F1 score performance metrics. The obtained results of accuracy are 87.88 for linear SVM, 86.06 for radial basis, 75.76 for sigmoid SVM, and 75.15 for polynomial SVM, respectively. The results show that for our experiments, Linear SVM performs better than radial, polynomial and sigmoid kernels. This is because projecting the data onto higher dimensions is not required as data exhibits linear properties confirmed by White Neural Network (WNN) test for nonlinearity.