AI-Augmented Medical Education: Transforming ICU Mortality * Length of Stay Prediction

Abstract

This research-to-practice full paper describes a dual application of artificial intelligence (AI) in both clinical prediction and medical education by developing machine learning (ML) models to predict ICU mortality and length of stay (LOS). Traditional scoring systems such as APACHE and SOFA are limited by their static nature and lack adaptability in real-time intensive care and instructional settings. Using the MIMIC-IV dataset, we trained and evaluated five ML models Random Forest, Logistic Regression, XGBoost, K-Nearest Neighbors, and Naive Bayes based on classification and error metrics. XGBoost demonstrated the best balance of performance and interpretability, achieving 95.50 % accuracy, strong recall (0.86), and the lowest scaled MAPE (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$134.93 \times 10^{12}$</tex>). To enhance educational value, SHAP explanations were integrated to help learners interpret feature importance and understand clinical reasoning behind predictions. These visual, data-driven insights align with cognitive apprenticeship and experiential learning theories, supporting competency-based education. Additionally, we propose a real-time educational dashboard that incorporates these interpretable models for simulationbased training environments. The findings suggest that interpretable AI models can both improve ICU outcome prediction and transform medical training by offering personalized, feedback-rich learning experiences.

Ähnliche Arbeiten