Predicting Emergency Department Mortality Risk Using Machine Learning Algorithms: The Case of Yekatit 12 Hospital Medical College

Background : The increasing burden of emergency department (ED) mortality in resource-constrained settings, such as Ethiopia, underscores the need for innovative approaches to enhance clinical decision-making. Machine learning (ML) shows promise in predicting mortality risk from electronic medical record (EMR) data, despite limited adoption in low- and middle-income countries (LMICs). Aim : This study aimed to curate EMR data and develop an optimized ML model for predicting mortality risk and support the clinical decision in the emergency department, at Yekatit 12 Hospital Medical College, Addis Ababa, Ethiopia. Methods : An experimental research design was employed, incorporating retrospective two-year cohort data from the EMR (October 2022–September 2024). Preprocessing involved data cleaning, transformation, feature engineering, and text mining to extract insights from unstructured records. Four ML models were trained and compared. Class imbalance was addressed using Synthetic Minority Oversampling Technique (SMOTE), undersampling, and hybrid methods. Hyperparameters were tuned via RandomizedSearchCV and GridSearchCV. Performance was evaluated with F1-score,  recall, precision, and Area Under the Receiver Operating Characteristic Curve (AUC-ROC). Results and Conclusion: The optimized XGBoost model achieved the best performance (F1-score of 0.97, precision of 0.95, recall of 0.99 for the alive class and F1-score of 0.68, precision of 0.89, recall of 0.55 for the minority class with AUC-ROC of 0.80). The study demonstrates that an optimized XGBoost model, leveraging routine EMR data, offers a scalable tool for mortality risk prediction in LMIC’s ED. Further improvements may require advanced imbalance handling and feature engineering methods.