Diabetic Prediction Using Confidence SMOTE Density Based Support Vector Machine

Diabetes is a chronic disease that affects millions of people in the world. In this work, we propose a confident version of the DBSVM classification method for the early detection and prediction of diabetes. The proposed method, called SMOTE Density Based Support Vector Machine (SDB-SVM) considers unbalanced data sets, we are going to present a way of handling unbalanced data sets by resampling methods (undersampling and oversampling). We will implement DB-SVM to our data set to clean it, then we will apply SMOTE to balance the classes well, then we will calculate the confidence degree of each synthetic point. Finally, we classified our new data according to the different classification methods. We compare our method to other known classifiers on a PIMA unbalanced diabetes data set and the Germany data set. The experimental results show the efficiency of the proposed algorithm.