Comparative Case Study: Machine Learning Approaches for Predicting Hospital-Acquired Infections (HAIs) in ICU Patients

Hospital-acquired infections (HAIs) are infections that develop during a patient’s hospitalization and are a significant clinical problem in intensive care units (ICUs). HAIs increase the risk of patient mortality, extend hospital stays, and can, in turn, drive up costs for patients and insurers, as well as damage a hospital’s reputation. This capstone project applies machine learning techniques to structured ICU data to explore how predictive models can support early detection of patients at elevated risk for HAIs. The dataset includes 29,604 ICU stays drawn from a larger cohort of 65,000 patient records, of which 2,207 (7.46%) were associated with an HAI. After data preparation and feature engineering, four machine learning models of logistic regression, random forests, XGBoost, and artificial neural networks were developed and evaluated to compare their predictive performance. The results of this project highlight the potential of machine learning to provide actionable insights that can improve infection prevention strategies and support proactive clinical decision-making in ICU settings.