Machine Learning approach for Sepsis Prediction in patients: A Comparative Study

Sepsis and multi-organ dysfunction syndrome (MODS) are critical clinical syndromes that need to be predicted in time to provide early clinical intervention. In this work, six machine learning models were trained to predict patient mortality: Decision Tree, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Logistic Regression, Random Forest, and XGBoost. An organized clinical dataset was pre-processed with feature selection and normalization, and then applied to train and test the models. Accuracy, Precision, Recall, and F1-score were used as evaluation metrics. Logistic Regression was the most successful overall classifier, with the best results, and KNN and XGBoost showed similar, balanced results. Conversely, Random Forest was overfitting, and Decision Tree had poor generalization. These results show that classical machine learning models, specifically Logistic Regression, KNN, and XGBoost, may offer balanced and consistent performance in sepsis mortality prediction, but the selection and tuning of models should be carefully considered for applicability in clinical contexts.