Applying machine learning algorithms and explainable AI to predict stunting and identify determinants among children under five in Ethiopia

One significant public health issue that impedes children’s growth and development is stunting. It is still very common in low-income nations. Reducing the impact of stunting requires early detection and determination of its factors. The purpose of this study was to forecast stunting in Ethiopian children under five and pinpoint its major contributing factors. This study used secondary data from the Ethiopia Demographic and Health Survey (EDHS). Machine learning algorithms, including logistic regression, random forest, decision trees, and XGBoost, were applied to identify determinants of stunting. Model performance was evaluated using accuracy, precision, recall, F1 score, balanced accuracy, Matthews Correlation Coefficient (MCC), and AUROC. The experimental findings revealed that the random forest model demonstrated robust predictive performance for stunting, achieving an accuracy of 95.3 ± 0.6%, precision of 93.9 ± 0.8%, recall of 96.6 ± 0.9%, F1-score of 90.5 ± 1.1%, balanced accuracy of 93.2 ± 0.7%, MCC of 0.90 ± 0.01, and AUC of 98.3 ± 0.5% across fivefold cross-validation, indicating high reliability and stability of the predictions. The most important predictors of stunting were underweight status, region, household wealth status, child’s age, mother’s age, place of delivery, birth order, maternal education, source of drinking water, and child’s sex. The Random Forest model proved to be the most successful in predicting stunting, according to the study. The frequency of stunting in low-income nations can be significantly decreased by enhancing maternal education, institutional delivery services, and availability to clean drinking water. To reduce childhood stunting, policymakers and program implementers should collaborate with important sectors like education, health, and agriculture.