Prediction of Melanoma Skin Cancer Using Veritable Support Vector Machine

Abstract

Classification of Medical Image is a superior technique for computer-aided diagnostic systems (CAD) towards skin cancer where it grows in fast manner than other diseases. Skin diseases are tolerable at sometimes and intolerable in many times. The typical approaches used to detect skin diseases are mainly based on characteristics, color, texture and their combinations. Primary objective of this research is to analyze the issues and challenges present in detecting the skin cancer, and to propose a novel classifier namely veritable support vector machine (VSVM) that can assist doctor to diagnosis the presence of skin cancer at early stage. VSVM includes the following steps: (i) train VSVM under supervised learning methodology to classify the medical image’s raw pixels using feature vectors that have high-ranking, (ii) derive a set of typical features that are based on information available in the history of medical images, (iii) construction of effective model that can combine different feature groups. VSVM focuses to classify the images that have the presence of skin cancer. VSVM provide an efficient way to build a classification model that can use raw pixels of medical image to construct the best classifier. However, since medical images are high in resolution and data sets are low, in-depth learning models are cost-effective and have minimum models. Matlab R2019b used to evaluate the performance of VSVM. The two datasets used for the evaluation of VSVM against SVM are HIS2828 and ISIC2017. Results indicate that VSVM has enhanced performance than SVM in terms of specificity, sensitivity, precision, recall, f-measure and classification accuracy. Comprehensive analysis of results indicates that the proposed classifier VSVM has increased performance in detecting the skin cancer and assist doctors than the previous classifier SVM.