ML-Powered ICU Mortality Prediction for Diabetic Patients

Abstract

Diabetes mellitus is one of the most important causes of mortality globally, particularly for critically ill patients undergoing treatment in ICUs. This study aims to enhance mortality prediction among diabetic ICU patients using advanced machine learning (ML) models. We tested several ML algorithms using a comprehensive dataset from the MIMIC III database, including Logistic Regression, Decision Tree, Random Forest, Support Vector Machine, and Multilayer Perceptron and compared their performances. The Random Forest model achieved the highest performance, with an AUC of 0.98, proving its effectiveness in managing complex datasets. Our models incorporate novel features such as patient demographics, lab results, and comorbidity indices, offering superior predictive power. This study highlights the critical role of ML in improving patient care by enabling timely interventions for high-risk ICU patients. Future research will focus on integrating real-time clinical data and refining the models to further enhance predictive accuracy.