CHALLENGES OF ARTIFICIAL INTELLIGENCE IN CLINICAL SEVERITY PROGNOSIS: A DECISION SUPPORT MODEL BASED ON HEMATOLOGICAL DATA — AN APPLIED STUDY ON COVID-19

This study proposes a decision-support model for COVID-19 severity prognosis using hematological data and machine learning methods. A retrospective dataset with 1,150 patients and 20 hematological predictors was analyzed after preprocessing, severity labeling, class distribution assessment, and stratified train-test splitting. Four models were evaluated: Logistic Regression, Support Vector Machine, XGBoost, and a Compact Neural Network. Model performance was assessed using accuracy, balanced accuracy, precision, sensitivity, specificity, F1-score, ROC AUC, average precision, and F2-score. Logistic Regression showed the most consistent performance, while threshold optimization improved sensitivity for severe cases, reducing false negatives at the cost of more false positives. The results indicate that hematological data provide limited but useful predictive signal for severity screening. The proposed approach is not intended to replace clinical judgment, but to support triage and prioritization decisions under class imbalance.