Comparison of Machine Learning Classifiers on Low Birth Weight in Indonesia: Study from Indonesia Demographic and Health Survey

The prevalence of low birth weight has increased, from 5.4% in 2007 to 6.2% in 2017. The purpose of this study is to examine the classification of infants' birth weight Indonesia, relies on 2017 Demographic and Health Survey. Machine learning classifiers, such as logistic regression, decision tree, naïve Bayes, random forest, support vector machine, and neural network, were used to model the classification of birth weight infants. This study also employed three scenarios of training-testing on the dataset. The training scenario giving the best performance was 90% ratio. In terms of accuracy, the best machine learning model on training dataset is the support vector machine. For testing dataset, random forest produces the highest accuracy for classifying the birth weight of infants in Indonesia, for the 10% and 30% scheme. The difference lies in 20% scheme where the highest accuracy was produced by decision tree. In whole testing scenarios, naïve Bayes produced the highest sensitivity, while the highest specificity is achieved by neural network. Considering the accuracy yielded by random forest model, mother's age at first birth, number of antenatal visits, number of family members, birth order (parity), and mother's occupation have a higher contribution to the class target of birth weight.