Comparison of linear discriminant analysis, support vector machines and naive bayes methods in the classification of neonatal hyperspectral signatures

Abstract

Hyperspectral imaging (HSI) is an imaging method that enables to obtain a large number of two-dimensional images in a wide wavelength range in the electromagnetic spectrum band. HSI has an important potential in biomedical applications as it provides diagnostic information about tissue physiology, morphology and composition. In this study, spectral signatures of unhealthy and healthy neonates were extracted using HSI method. Then, data statistics (minimum, maximum, mean, median and standard deviation) of these signatures were obtained and classified with Linear Discriminant Analysis, Support Vector Machines and Naive Bayes Methods. After the classification processes, the best result was achieved using Linear Discriminant Analysis with 90.63% accuracy, 87.50% sensitivity and 93.75% specificity. © 2021 IEEE.