Enhancing Diabetes Hospital Stay Predictions through Interpretability

Diabetes Mellitus is a prevalent chronic disease with significant public health implications, often leading to hospitalizations due to complications. Accurate prediction of the length of hospital stay (LOS) is essential for effective patient management and resource allocation. This study focuses on enhancing LOS predictions for diabetes patients using machine learning models, specifically Random Forest and Feed-forward Neural Networks. Utilizing a dataset of over 70,000 patient records from more than 120 U.S. hospitals, key variables such as demographics, admission types, and lab results were analyzed. The findings indicate that incorporating Shapley values improved model interpretability and bolstered confidence in predictive outcomes via enhancing the accuracy and precision of LOS predictions.

CCS Concepts ? Machine learning ? Healthcare applications ? Interpretability in AI