Use of artificial neural networks for emergencies prediction at Hospital Universitari de Girona Dr. Josep Trueta
Document typeBachelor thesis
Rights accessOpen Access
The motivation of the project is to model and predict the volume of arrivals at the emergency department (ED) of a general hospital. The process consists of complex linear and nonlinear patterns together. Those types of temporal series are tough to solve efficiently using Box-Jenkins methods (ARIMA models) due its high stochastic behaviour and nonlinearity. Once the time series analysis is discarded owing the bad results obtained, and in order to change the approach of the task, artificial neural networks (ANN) are chosen to solve the problem. This methodology offers a whole new perspective of study, enabling the use of algorithms in a high tight time constraint in order to predict intraday information such as the arrivals expected to occur in the afternoon using morning information. The objective is to program a plain applicative, able to extract the data needed (endogenous and exogenous variables), compute the ANN algorithm and finally show the relevant results in order to help improving the human and material resources management in the area of emergencies. As a fundamental part of the project, the best methodology to work with ANN algorithms is seek in order to settle an accurate approach for future studies.