Find in Library

Development of computer aided system to diagnose dermatological disorders works as a second opinion when skin diseases have very little differences in clinical features. Support Vector Machine (SVM) is a good classifier for non linear data with appropriate choice of kernels. Generally, Positive (semi) Definite (PSD) kernels called Mercer’s kernels are used in SVM. Mercer’s condition is the traditional requirement for classical kernel methods like SVM. But, for many empirical data indefinite kernels can give better result. In this study we use SVM with a novel kernel (Modified Gaussian Kernel) which is Indefinite (ID) kernel to diagnose skin disorders. We also investigate various distance substitution kernels to diagnose skin disorders and determine Eigen values of the Gram matrices obtained from two dermatological data sets under study to discuss their definiteness property. Results show that our proposed modified Gaussian kernel gives good classification accuracy to diagnose dermatological disorders.