Project Overview
Determining which patients are at greatest risk early in an ICU stay is critical for guiding timely clinical interventions. In this capstone research project for my Data Science master’s degree, I integrated multiple data sources from the MIMIC-IV database to build predictive machine learning and deep learning models for in-hospital mortality in HF ICU patients, achieving an overall accuracy of 89%. By adapting to a highly imbalanced dataset using techniques like focal loss and threshold optimization, the approach not only demonstrates the capacity for early mortality detection but also accounts for the real-world challenges inherent in making data-driven clinical decisions.
If the embedded PDF does not display, you can download the summary report here.