A Comparative Study of Machine Learning Algorithms in Predicting Acute Respiratory Infection of Under-Five Children in Bangladesh Using Imbalanced Data

This study inspects the occurrence and indicators of acute Respiratory Infection (ARI) among under-five children in Bangladesh using data from the 2017 and 2022 Bangladesh Demographic and Health Survey (BDHS), and it incorporates 11157 surviving children after excluding visitors and children aged 5 or older. ARI was pointed out in 2.5% of children. SMOTE was used to make the dataset balanced. Ten machine learning algorithms, including Artificial Neural Networks, AdaBoost, Classification and Regression Trees, K-nearest Neighbors, Logistic Regression, Linear Discriminant Analysis, Support Vector Machines, Random Forests, Naive Bayes, and XGBoost were employed to model ARI incidence. Data was partitioned using the 80-20 rule. A 10-fold CV was implemented for parameter tuning. XGBoost obtained the highest accuracy (91%), followed by SVM (90.1%) and Random Forest (88.9%). XGBoost also showed the highest sensitivity (87.8%). The SVM had the best ROC at 0.963, indicating the superior overall performance of these models. In contrast, Naive Bayes demonstrated the poorest performance with an accuracy of 68.9% and a specificity of 56.4%. The Kappa statistics were obtained by using XGBoost with the highest value of 0.82. These outputs suggest that XGBoost and SVM are highly effective in predicting ARI among children under five years.