M.S. AAI Capstone Chronicles 2024

Introduction This project addresses the critical challenge of predicting sepsis onset in ICU patients by applying machine learning techniques to clinical data that is commonly available in the ICU. Sepsis, a severe inflammatory response to infection, can progress quickly and is a leading cause of mortality in intensive care settings (Singer et al., 2016). Early detection of sepsis is essential, as timely medical intervention can drastically reduce mortality rates and improve patient outcomes (Fleischmann et al., 2016). Our goal is to develop a predictive model capable of accurately identifying the onset of sepsis at least six hours before clinical symptoms are observed, giving healthcare providers essential lead time for intervention. The model would be deployed as part of an early warning system for healthcare professionals in ICU environments, where rapid response to early indicators of sepsis can prevent the progression to septic shock and other severe complications. The dataset selected for this project is the Prediction of Sepsis dataset from the PhysioNet Computing in Cardiology Challenge 2019, which includes extensive time-series physiological data, such as vital signs, lab results, and demographic information critical to sepsis prediction (Reyna et al., 2019). In a live system, the model would process real-time data from ICU patient monitoring devices, allowing for continuous sepsis risk assessment and timely alerts for clinicians. The ultimate goal of this project was to develop a reliable, high-performing tool for early sepsis detection. By developing this predictive model, we aim to contribute to healthcare AI by offering a practical tool for ICU settings and advancing the understanding of Machine Learning (ML) and Deep Learning (DL) techniques for critical care applications. This project’s findings could help improve the use of predictive modeling in healthcare, especially for conditions

265