Improving Hospital Bed Management: Machine Learning for Predicting Length of Hospital Stay

Abstract

This study proposes a novel approach to managing short-term inpatient stays by predicting the length of stay (LOS) upon patient admission using machine learning (ML). This study demonstrates that ML algorithms (i.e., Random Forest, XGBoost, and Convolutional Neural Network can predict LOS with almost twice the accuracy compared to conventional physicians’ assessment. As a result of the higher LOS estimation accuracy, simulation results show a higher bed occupancy rate and lower patient waiting time. Overall, this study concludes that with better decision support system for short-term hospitalization management, a hospital can increase the number of short-term patient admissions or operate with fewer hospital beds.

Keyword: length of hospital stays, machine learning, hospital bed management, discrete event simulation