Icu Outcomes Prediction Using Optimized Extra Trees Classifier And Lasso-Based Feature Selection Publisher



Nematollahi MA ; Maftoun M ; Khademi M ; Sori AA ; Atashi A ; Alizadehsani R ; Moosaei H
Source: Lecture Notes in Computer Science Published:2026

Abstract

Accurately predicting outcome among ICU patients is essential for enhancing clinical decision-making and optimizing healthcare resource management. This study leverages machine learning techniques to improve prediction accuracy in this critical domain. Comprehensive data preprocessing steps were undertaken, including handling missing values, normalizing data, and addressing class imbalances. By utilizing LASSO, 12 key predictive features were identified. Eleven machine learning models, including traditional methods such as Gaussian Naive Bayes, Logistic Regression, and Support Vector Machines, as well as four advanced deep learning architectures like CNN-BiLSTM and hybrid models, were evaluated. Hyperparameters were optimized using Bayesian optimization to achieve peak model performance. Among these, the Extra Trees classifier outperformed others, achieving an AUC of 96.61%, accuracy of 91.80%, sensitivity of 92.81%, specificity of 91.03%, precision of 90.79%, and an F1 score of 91.91%. Cohen’s Kappa and the Matthews Correlation Coefficient (MCC) both reached 83.62%. The findings highlight the potential of advanced machine learning models to significantly enhance outcomes prediction, thereby supporting better clinical decision-making and improving patient outcomes. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.