An explainable AI framework for predicting respiratory tract infection using symptoms and multidimensional risk factors in resource-scarce settings

Respiratory Tract Infections (RTIs) pose significant health challenges in resource-scarce settings, where access to diagnostic tools is limited. This study presents an explainable artificial intelligence (XAI) framework for predicting RTIs using patient symptoms and multidimensional risk factors. Leveraging a dataset of 4,870 patient records provided by the New Frontiers in Research Fund (NFRF), we addressed class imbalance, with 1,756 RTI-positive and 3,112 RTI-negative cases, using the Synthetic Minority Over-sampling Technique (SMOTE). This resulted in a balanced dataset of 6,258 instances (3,129 positive and 3,129 negative). We implemented and evaluated three traditional machine learning models: Random Forest, XGBoost, and Decision Tree. To enhance model transparency and support clinical trust, we integrated Local Interpretable Model-Agnostic Explanations (LIME) for feature-level interpretability. Among the models, XGBoost achieved the highest AUC-ROC of 0.8876, followed closely by Random Forest (0.8823), while Decision Tree performed lowest (0.7922). LIME explanations consistently identified Dizziness (DIZ) and Shock (SHK) as the most influential predictors of RTI, reflecting symptoms associated with severe infections. Fever was also highlighted, underscoring alignment with clinical understanding. This framework not only achieves high predictive performance but also promotes interpretability, making it suitable for real-world deployment in healthcare systems with limited resources.