A COMPARISON OF ARTIFICIAL NEURAL NETWORK AND NAIVE BAYES CLASSIFICATION USING UNBALANCED DATA HANDLING

Abstract

Classification is a supervised learning method that predicts the class of objects whose labels are unknown. Classification in machine learning will produce good performance if it has a balanced data class on the response variable. Therefore, unbalanced classification is a problem that must be taken seriously. This study will handle unbalanced data using the Synthetic Minority Over-Sampling Technique (SMOTE). The classification methods that are quite popular are the Naïve Bayes Classifier (NB) and the Resilient Backpropagation Artificial Neural Network (Rprop-ANN). The data used comes from the Health Nutrition Research and Development Agency (Balitbangkes) which consists of 2499 observations. This study examines the use of NB and ANN using the SMOTE method to classify the incidence of anemia in young women in Indonesia. Modeling is done on 80% of training data and predictions on 20% of test data. The analysis shows that SMOTE can perform better than not handling unbalanced data. Based on the results of the study, the best method for predicting the incidence of anemia is the Naïve Bayes method, with the sensitivity value of 82%.