Loading...
Using Machine Learning Models for Inpatient Bed Demand Forecasting in the UAE Hospitals
Sheiko, Rahaf Abdulla
Sheiko, Rahaf Abdulla
Date
2024-07
Author
Advisor
Type
Thesis
Degree
Citations
Altmetric:
Files
35.232-2024.77a Rahaf Abdulla Sheiko.pdf
Adobe PDF, 1.31 MB
- Embargoed until 2027-04-14
Description
A Master of Science thesis in Engineering Systems Management by Rahaf Abdulla Sheiko entitled, “Using Machine Learning Models for Inpatient Bed Demand Forecasting in the UAE Hospitals”, submitted in July 2024. Thesis advisor is Dr. Rami Afif As’ad and thesis co-advisor is Dr. Hussam Alshraideh. Soft copy is available (Thesis, Completion Certificate, Approval Signatures, and AUS Archives Consent Form).
Abstract
Healthcare is a crucial global sector that impacts the physical, mental, and social wellbeing of people. Accurate healthcare forecasts help bridge the gap between supply and demand of healthcare resources, ensuring accessible services for patients. On the other hand, lack of accurate healthcare forecasts can lead to several issues including increased transfer times, delays in elective surgeries, and increased hospital safety incidents. Accurate demand forecasting is a challenging research problem, with most studies focusing on specific health conditions or aggregate scenarios such as Emergency Department (ED) scenario. This thesis draws focus on planning the inpatient bed demand in UAE hospitals. This thesis proposes the deployment of different machine learning models to accurately predict the daily forecasts of inpatient bed demand and accordingly assist with resource planning for the hospitals. The proposed models will be tested on the healthcare data set including inpatients records from 2018 to 2021, collected from Emirates Healthcare Services (EHS) hospitals. The proposed models will be assessed based on predefined metrics including the Squared Correlation and the Root Mean Squared Error (RMSE). Then, a Multi Criteria Decision Making (MCDM) tool is used to select the proper model based on factors including accuracy, simplicity, interpretability, computational time, and implementational feasibility, combining the performance metrics and the experts’ input collected through a survey, which is the novelty of the work. As for the results, the XGB and KNN were the best performing when assessing the models in terms of RMSE and Squared Correlation, achieving an RMSE of 1.816 beds/day and 2.486 beds/day respectively, and a Squared Correlation of 0.997 and 0.993 respectively. The AHP was then used to incorporate the model’s performance metrics with the experts input, Random Forests were indeed the best performing achieving a value of 0.24 in the AHP and 0.851 and 12.990 for Squared Correlation and RSME respectively followed by KNN and XGB.