Comparison of performance of different statistical machine learning algorithms to predict Child Survival status in North-Eastern states of India

Abstract

In today's modern world, reducing child mortality rates are the primary health priority in developing countries like India. Identifying determinant factors of under-five child mortality will support notifying parents, health policymakers, and practitioners to adopt effective and efficient actions to improve child health, altering the children's future. This study aims to develop different predictive models to classify children under five (alive or dead) survival status. A logistic regression algorithm is employed to extricate novel factors associated with child death that can serve as targets for intervention. The four commonly used machine learning algorithms- Logistic Regression (LR), K-Nearest Neighbors (KNN), Naïve Bayes (NB), and Support Vector Machine (SVM) are employed to design a predictive model of child survival status, using National Family Health Survey (NFHS-4) 2014-15 data. This work contributes to exploring the possibilities of using supervised learning algorithms to classify children effectively. The predictive performance measures used for classification are precision, recall, f1-measure, ROC (receiver operating characteristics curve), and AUC (area under ROC curve). Our findings reveal Logistic Regression makes the best classification accuracy.