The success of hospitals in treating patients and staying in business relies on their efficient use of resources. In particular, the utilization of hospital beds is a critical concern, since over-crowding will result in delays or transfers of patients, and under-utilization will result in lost opportunity to treat patients and generate profit. To this end, hospital decision makers must have reliable forecasts of patient demand and bed availability. The objective of this thesis was to create a general method to forecast the availability of hospital beds in the short term, up to 2 days into the future. Specifically, this thesis employed a computer simulation model of the hospital and a time-dependent neural network to learn from the simulated model and forecast the availability of beds. The computer simulation model was found to be well suited to the task of describing a general hospital system and creating training data for a neural network. The neural network was found to provide accurate performance in predicting bed availability in the short term. The network incorporated the effect of time explicitly to capture the non-stationary behavior of hospital systems. These findings have a number of implications that will be discussed.

Library of Congress Subject Headings

Hospital size--Forecasting; Hospital size--Computer simulation; Neural networks (Computer science)

Publication Date


Document Type


Student Type


Degree Name

Industrial and Systems Engineering (MS)

Department, Program, or Center

Industrial and Systems Engineering (KGCOE)


Michael Kuhl

Advisor/Committee Member

Moises Sudit

Advisor/Committee Member

Elisabeth Hager


Physical copy available from RIT's Wallace Library at RA963 .D36 2004


RIT – Main Campus