Predicting patients discharge and Optimizing hospital bed management using AI models

Yovan, Annuncia Marena (2025) Predicting patients discharge and Optimizing hospital bed management using AI models. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (416kB) | Preview

Abstract

Healthcare systems worldwide face immense pressure from rising patient numbers, leading to hospital overcrowding, prolonged wait times, and inefficient resource allocation. Effective hospital bed management is critical to mitigating these challenges, yet it is often hampered by the difficulty in accurately predicting patient discharge times. This research project develops an integrated framework that leverages artificial intelligence to address this dual problem. The primary objective is to predict patient length of stay (LOS) with high accuracy and then use these predictions to optimise bed allocation. Utilising the 2010 New York State Hospital Inpatient Discharge data, this research applies and assesses a group of machine learning models. Enhanced gradient boosting methodologies, CatBoost and LightGBM, are evaluated relative to baseline models including models such as, Linear Regression, Random Forest, and Support Vector Machines. The results show the CatBost had the best performance and discovered a Root Mean Squared Error (RMSE) of 2.67 (using a sample of the data). To improve model interpretability, SHAP (SHapley Additive exPlanations) was used to understand the predictions from the ML models. The SHAP analysis revealed diagnosis codes, total charges, and severity of illness as the most important predictors of LOS. The discharge prediction from CatBoost was then embedded into a Mixed-Integer Linear Programming (MILP) model to assign patient discharges to available beds across several hospital wards. The optimisation MILP model effectively assigned patient discharges to available beds while respecting ward capacity and presents a promising operational planning method. The contribution from this project consists of a full, data-driven prediction and assessment framework, enhancing prediction accuracy and embedding an actionable framework for optimising intervention in hospital resource management systems, aimed at improving patient flow and care quality.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Hamill, David UNSPECIFIED
Subjects:	Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence R Medicine > Healthcare Industry Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	03 Jul 2026 11:42
Last Modified:	03 Jul 2026 11:42
URI:	https://norma.ncirl.ie/id/eprint/9469

Actions (login required)

View Item