A Comparative Analysis of Machine Learning Models for the Classification of Heart Failure Patients in the Intensive Care Unit

Abstract

Background: Heart failure is the leading cause of death globally over the last several decades. This raises the necessity of timely, accurate, and prudent methods for establishing an early diagnosis and implementing timely illness care. Objective: This study aims to develop and validate a classification model for the patients admitted to the Intensive Care Unit (ICU) with heart failure, using various machine learning models applied to the MIMIC (Medical Information Mart for Intensive Care)-III database. Method: A retrospective cohort study was conducted using data extracted from the MIMIC-III database. Machine learning models: Logistic Regression, K-Nearest Neighbor (KNN), Random Forest, Decision Tree, Naïve Bayes, AdaBoost, and XGBoost were utilized to construct the predictive model. The dataset has been preprocessed in two different manners. The study included 1,177 patients with heart failure, selected according to specific inclusion/exclusion criteria and admitted to the ICU. Result: At the end of the study, the most effective model for predicting patients who survived was Logistic Regression, with an accuracy of 0.9025, sensitivity of 0.9763, precision of 0.9196, and F1-score of 0.9471. Conclusion: Classification of the patients into those who survived or could not survive due to heart failure was the primary measure, with various clinical and demographic variables used as predictors.