A new approach to modelling and forecasting monthly overnights in the Northern ...

• Autor(es): Fernandes, Paula O. ; Teixeira, João Paulo
• Data: 2007
• Identificador Persistente: http://hdl.handle.net/10198/1021
• Origem: Biblioteca Digital do IPB
• Assunto(s): Artificial neural networks; Training; Backpropagation; Forecasting
• Tipo de Documento: Documento de conferência
• Idioma: Inglês

Descrição

The need to analyze the main factors determining the evolution of demand within the tourism sector, which is the driving force of the whole tourism activity, and the importance that forecasting has in this domain, may be justified by the fact that the tourism sector plays a significant role in the economy of Portugal and its regions because of the large number of people employed directly and indirectly, and also because of its ability to bring in currency that reflects in different sector of economic activity. Although tourism is less developed in the North of Portugal than in other regions of the country, it is essential to comprehend this phenomenon in order to empower local economic agents to carry out strategic measures to maximize profits from newly emerging situations. The objective of the present research is to quantify national and international tourism flows by developing (mathematical) models and applying them to sensitivity studies in order to predict demand. This work provides a deeper understanding of the tourism sector in Northern Portugal and contributes to already existing econometric studies by using the Artificial Neural Networks methodology. This work's focus is on the treatment, analysis, and modelling of time series representing “Monthly Guest Nights in Hotels” in Northern Portugal recorded between January 1987 and December 2003. This was achieved through a study of the reference time series whose past values were known and whose objective was to obtain a model that better predicts the behaviour of the time series under study. The model used 6 neurons in the hidden layer with the logistic activation function and was trained using the Resilient Backpropagation algorithm (a variation of backpropagation algorithm). Each time series forecast depended on 12 preceding values. The obtained model yielded acceptable goodness of fit and statistical properties and is therefore adequate for the modelling and prediction of the reference time series.