Predicting breastfeeding practice of Nigerian child using machine learning and deep learning algorithms

Abstract

The accessibility and rapid growth of data with the conspicuous rise in hardware technologies have led to the development of new studies in distributed and deep learning. This work reports the prediction of breastfeeding practice of a child using machine and deep learning algorithms. Despite that exclusive breastfeeding reduces the child mortality rate caused by pneumonia and diarrhea and other benefits attached to it the percentage of mothers who practice it in Nigeria still fall short to 29% in 2018 according to Nigeria Demographic Health Survey (NDHS) compare to the global target of 70% by 2030. The aim of the study is to develop a model for the prediction of breastfeeding practice in Nigeria. This study adopted Sample, Explore, Modify, Model and Access (SEMMA) data mining methodology using 2018 NDHS dataset. Experiments were performed with ML algorithms (RF, J48, JRIP, and SVM) built with WEKA software as well as with DNN algorithm using python. Some control parameters were applied to configure the DNN model while considering the number of layers, neurons within each layer, activation function for each layer, ADAM algorithm, epochs (iterations), learning rate and percentage split of dataset between training and testing subsets. The performance of the DNN model in predicting breastfeeding practice was evaluated and compared with previous study that used ML models only. It was found that DL has a better child prediction of breastfeeding practice than ML models with accuracy of 97.9%. This could serve as a supporting tool for healthcare practitioners to support breastfeeding practice in Nigeria.