Performance of SMOTE in a random forest and naive Bayes classifier for imbalanced Hepatitis-B vaccination status

Abstract

The alternative approach to predict the classification of HB vaccination status is using a machine learning approach such as random forest and naive Bayes classifier. However, for imbalanced classification, the algorithms are biased towards the majority class. To increase the accurate prediction of the classifier, we consider the Synthetic Minority Oversampling Technique (SMOTE) to have more balanced data. The purpose of this study was to compare the performance of SMOTE in the random forest and naive Bayes classifier for imbalanced Hepatitis-B vaccination data. The study used the National Socio-Economic Survey data in 2017 for Aceh province with 2264 cases and 14 variables. The results show that the application of SMOTE in the random forest and naive Bayes classifier improves the accuracy of identification of Hepatitis-B non-vaccination status by 30.08% and 26.09%, respectively, compared to non-SMOTE. Random forest with SMOTE is the best model for classification HB vaccination status. The most important factors that influence the Hepatitis-B vaccination status of Aceh province are the mother’s last education, mother’s occupation, father’s occupation, father’s previous education, and the number of health facilities.