A New Approach of Hybrid Sampling SMOTE and ENN to the Accuracy of Machine Learning Methods on Unbalanced Diabetes Disease Data

Abstract

The performance of machine learning methods in disease classification is affected by the quality of the dataset, one of which is unbalanced data. One example of health data that has unbalanced data is diabetes disease data. If unbalanced data is not addressed, it can affect the performance of the classification method. Therefore, this research proposed the SMOTE-ENN approach to improving the performance of the Support Vector Machine (SVM) and Random Forest classification methods for diabetes disease prediction. The methods used in this research were SVM and Random Forest classification methods with SMOTE-ENN. The SMOTE-ENN method was used to balance the diabetes data and remove noise data adjacent to the majority and minority classes. Data that has been balanced was predicted using SVM and Random Forest methods based on the division of training and testing data with 10-fold cross-validation. The results of this study were Random Forest method with SMOTE-ENN got the best performance compared to the SVM method, such as accuracy of 95.8%, sensitivity of 98.3%, and specificity of 92.5%. In addition, the proposed method approach (Random Forest with SMOTE-ENN) also obtained the best accuracy compared to previous studies referenced. Thus, the proposed method can be adopted to predict diabetes in a health application.