Comparison of Under-sampling and Over-sampling Techniques in Diabetic Mellitus (DM) Patient Data Classification by Using Naive Bayes Classifier (NBC)

Abstract

Imbalance dataset is a big problem inside a classification process. Most of the classification algorithms tend to classify the majority instances and ignore the minority ones. It can cause the misclassification of the minority instances and make the precision and recall of this minority data become low. In order to resolve this kind of problem there will be done both undersampling and oversampling process to make the dataset balance. In this proposed research there will be used undersampling and oversampling techniques to balance the number of majority and minority instances from diabetic patient data. The other techniques used in this research are backward greedy stepwise for features selection and Naive Bayes Classifier (NBC) for data classification. The conclusion, oversampling techniques give significantly higher precision and recall than oversampling, although the accuracy fairly equal.