Enhancing ICU Patient Outcomes Trough Machine Learning and Ensemble Technique

Abstract

In order to identify the mortality risk and estimate the death rate of ICU patients, machine learning (ML) and ensemble learning approaches are utilized to analyse a variety of patient data and provide an accurate prognosis. The mortality rate of ICU patients nowadays is very high. If we can identify the reason as early as possible then we can start diagnosis as early as possible. First, relevant attributes such as test results, symptoms, and demographic data are taken out of patient files. ML algorithms like logistic regression, decision trees, and support vector machines classify patients into low- and high-risk categories during the mortality risk identification stage. Through model aggregation, ensemble techniques like random forests and gradient boosting improve predictive performance. Regression models such as ridge regression, neural networks, and linear regression evaluate the probability of death within given time frames to predict mortality rates. These estimates are then improved using ensemble learning strategies like stacking or bagging. This abstraction, which improves patient outcomes in the ICU, enables healthcare workers to quickly identify at-risk patients and make well-informed decisions through careful feature engineering, model hyperparameter tuning, and cross-validation. The best accuracy comes with logistic regression (90%) and 2nd highest accuracy comes with both random forest and KNN which is 89%. The results of xgboost and adaboost are 87.07% and 84%