Comparison of Machine Learning Techniques for Determinants of Low Birth Weight Infants in Indonesia

Low birth weight accounts for 35.15 percent of all infant fatalities in Indonesia, making it the biggest cause of newborn death in 2020. The purpose of this study is to figure out the determinants influencing low birth weight infants in Indonesia. The Demographic and Health Survey dataset was used for this study in both 2012 and 2017. Machine learning techniques, including multiple linear regression, regression tree, Random Forest, Support Vector Regression, and neural network (NN), were used to model low birth weight infants. This study also employed two scenarios of training-testing on both datasets. Low birth weight, as modeled by regression trees and neural networks, could account for almost 20% of the variation in low birth weight infants in the 2012 dataset. A similar pattern was figured out in the 2017 dataset. When compared to other methods, the neural network method produces the smallest root mean square error (RMSE) on both training datasets. Using NN as the preferred approach, the determinants having the greatest contribution to the low birth weight of infants in the 2012 dataset are infants’ alive status, number of antenatal care visit, literacy index, and birth order. According to the 2017 dataset, the determinants with the greatest contribution on the low birth weight of infants are infants’ alive status, number of antenatal care visit, and literacy index. In terms of predictive performance as measured by RMSE, NN produced lower RMSE than the others.