Improving Prediction of Mortality in ICU via Fusion of SelectKBest with SMOTE Method and Extra Tree Classifier

The healthcare system’s increasing adoption of analytics and advanced computers demonstrates a growing tendency to develop a strong prognostication mechanism based on Artificial Intelligence (AI), utilizing technology to explore hidden relations between data and assessment in medical contexts. Hence, predicting the mortality of Intensive Care Unit (ICU) patients is a vital yet challenging task with significant implications for clinical decision-making in healthcare. This study aims to present an effective approach based on the AutoML framework for building Machine Learning (ML) models to predict mortality. PyCaret was applied as an AutoML framework in this study. ML approaches such as Extra Tree Classifier (ET), eXtreme Gradient Boosting (XGBoost), Random Forest (RF), Light Gradient Boosting Machine (LightGBM), Gradient Boosting Classifier (GBC), and Adaptive Boosting (AdaBoost), along with feature selection technique, SMOTE, and Auto Fine Tuned via Grid-Random search (GRS) by the assistance of PyCaret are employed to predict mortality. Results demonstrate that the hybrid (SelectKBest + SMOTE + ET) method achieves the highest AUROC of 93.54% for prediction, outperforming other models.