Background: The COVID-19 pandemic has posed immense challenges to healthcare systems worldwide, making it critical to identify predictors of patient mortality. Machine learning algorithms have shown great promise in predicting outcomes for COVID-19 patients based on clinical and laboratory data. This study aimed to evaluate the performance of three machine learning models—Random Forest, Support Vector Machine (SVM), and Logistic Regression—in predicting mortality among COVID-19 patients using clinical and laboratory data.
Methods: A retrospective cohort study was conducted on 2,500 COVID-19 patients admitted to three major hospitals in Tehran, Iran, between 2020 and 2021. Demographic, clinical, and laboratory data were collected. Machine learning models were trained to predict mortality, and their performance was evaluated using the area under the curve (AUC), sensitivity, specificity, and positive predictive value (PPV). Model validation was performed using 10-fold cross-validation and an 80-20 train-test split to ensure robustness and generalizability.
Results: The Random Forest model outperformed both SVM and Logistic Regression with an AUC of 83.3%, sensitivity of 63.0%, and specificity of 90.5%. Important predictors of mortality included age, gender, comorbidities (such as diabetes, ischemic heart disease, and cancer), ICU admission, and laboratory markers (e.g., ALT, white blood cell count, and creatinine levels).
Conclusion: The Random Forest algorithm demonstrated superior predictive performance compared to SVM and Logistic Regression in predicting mortality among COVID-19 patients. These findings suggest that machine learning models, particularly Random Forest, can be instrumental in identifying high-risk patients and supporting clinical decision-making.