期刊名称：International Journal of Advances in Soft Computing and Its Applications
印刷版ISSN：2074-8523
出版年度：2019
卷号：11
期号：3
页码：76-92
出版社：International Center for Scientific Research and Studies
摘要：Diabetic Retinopathy (DR) is one of the frequent comorbidities of diabetes worldwide. Diabetic eye screening has become a challenging task for ophthalmologist as they need to deal with large number of patients to be diagnosed, creating a need to develop tool that may help ophthalmologist to classify the severity of DR in order to establish an adequate therapy. Previous researchers have studied machine learning to propose an automatic DR classification using the clinical variables. However, it needs to be improvised especially in terms of accuracy. Hence, this paper aims to propose an optimal or near-optimal DR classifier using the Support Vector Machine with hyperparameter optimization. This study considered three classes of diabetic patients which were patients who do not have DR (NODR), patients with non-proliferative DR (NPDR) and patients with proliferative DR (PDR), instead of focusing only on two classes (NO DR, DR). The radial basis function, polynomial, sigmoid kernel and their respective hyperparameters were tested in this study in order to find the best kernel and combination of hyperparameters that can improve the performance of SVM. The results obtained show that SVM-radial kernel with cost value,64, 0.03 gives the best accuracy at 85.45%.
