Machine Learning Algorithms for understanding the determinants of under-five Mortality

Area of Science:

• Public Health
• Data Science
• Biostatistics

Background:

• Under-five mortality remains a critical global health and social development issue.
• Predictive modeling can aid in identifying at-risk populations and informing interventions.

Purpose of the Study:

• To evaluate the accuracy of various machine learning models in predicting under-five mortality.
• To identify significant factors associated with under-five mortality.

Main Methods:

• Utilized data from the National Family Health Survey (NFHS-IV) for Uttar Pradesh.
• Applied machine learning techniques including logistic regression, decision trees, random forest, Naïve Bayes, KNN, SVM, neural networks, and ridge classifier.
• Assessed model performance using metrics like accuracy, precision, recall, F1 score, Cohen's Kappa, and AUROC.

Main Results:

• The neural network model demonstrated superior predictive accuracy for under-five mortality (95.29%–95.96%).
• Logistic regression also performed well, with accuracy ranging from 94% to 95%.
• Key factors identified include number of living children, survival time, wealth index, child size at birth, mother's education, and birth order.

Conclusions:

• Neural networks offer a highly accurate approach for predicting under-five mortality.
• Machine learning models, including logistic regression, provide valuable tools for analyzing high-dimensional health data.
• Findings can inform targeted public health strategies to reduce child mortality.