Machine learning approach for predicting infant mortality in sub-saharan Africa

Objectives: Infant Mortality (IM) is a serious issue in Sub-Saharan Africa (sSA). Understanding the factors that affect infant health is crucial for saving lives. This study aimed to develop and evaluate high performance Machine Learning (ML) model that can effectively identify the risk of IM which will help to improve intervention plans or policies meant to lower IM in the area.

Study Design: This study is based on cross-sectional survey

Methods: This research utilizes data from the recent Demographic and Health Survey (DHS) across 30 sSA countries. It employs various ML such as Decision Tree, Bayesian Network, eXGBoost, Random Forest, Logistic Regression, Linear Discriminant Analysis, Super Vector Machine, Artificial Neural Network and K-Nearest Neighbors. The effectiveness of these algorithms is evaluated using metrics like accuracy, precision, recall, F1 score, and the Area Under the Receiver Operating Characteristics (AUROC). The model interpretability was attained by using SHAP values and Feature importance to quantify variable contributions.

Results: The descriptive results highlight the variations in IM across different sSA countries. By training 80% of the dataset and using the remaining 20% for testing, the eXGBoost algorithm outperformed others, achieving highest accuracy of 96.0%, a recall of 96.5%, a precision of 84.15%, an F1 score of 86.0%, and an AUROC of 0.984.  Based on SHAP value analysis key protective factors were high number of deceased children, low infant weight, first birth order, vaccination coverage, multiple births, pregnancy duration: and the risk factors were having postnatal checkups, narrow births interval, and low amenorrhea and malaria prevalence.

Conclusions: This study employs various ML approach and identified the key factors.  To reduce IM in sSA, interventions should focus on enhancing postnatal care, ensuring vaccination coverage, implementation of surveillance for mothers with high number of died children, managing multiple births, advocating for birth spacing, providing clinical management for birth interval, birth order, duration of amenorrhea and duration of pregnancy and implementing malaria prevention measures. These strategies can significantly enhance infant health outcomes in the sSA.

Keywords: Machine Learning, Infant Mortality, Ssa, Public Health.