Tourism Demand Forecasting Based on a Neuro-Fuzzy Model

1. Introduction

When looking at the history of Greece, the growth rates were extremely high and exceeded the rest of Europe for the two decades after 1950. This boosted the country’s economy and the further development of tourism. For the years post-1980, Greece has attracted more than 15 million tourists each year and was ranked in 21^st place for competitiveness in Europe by the World Economic Forum in 2011. Furthermore, in that year, travel and tourism contributed 16.5% to Greece’s GDP and 18.4% to employment.¹

However, the perplexity that is inherent in tourism forecasting makes it hard to predict. Tourism is easily affected by internal and external conditions such as political, financial or environmental factors. Nevertheless, an accurate forecast of tourism demand can be beneficial to a country and tourism-related industries, which will benefit through building tourism infrastructures, in making fundamental strategic decisions or in taking precautionary measures for potential threats and opportunities.

Martin and Witt (1989) were some of the first groups of authors to study tourism demand forecasting. They used the ANOVA and Scheffé tests and produced more accurate forecasts by adapting simple forecasting methods than econometric forecasts. The use of Artificial Intelligence (AI) based techniques in tourism though, started around the mid-1990s and has become more popular over recent years. One of the oldest studies to apply AI was that of Pattie & Snyder (1996) who used a back-propagation neural network model to forecast tourism demand in US national park systems. Tsaur et al. (1997) used an Analytic Hierarchy Process method and a fuzzy Multiple Criteria Decision-Making method to conduct the evaluation of tourist risk. Law & Au (1999) presented a new approach that uses a supervised feed-forward neural network model to forecast Japanese tourist arrivals in Hong Kong, preferable to multiple regressions. A later study by Law (2000) demonstrated a back-propagation learning approach that can give more accurate results in a neural network’s forecast in relation to tourism demand for non-linearly separable dependent variables. Song et al. (2003) evaluated the forecasting accuracy of six alternative econometric models in the context of the demand for international tourism in Denmark. Pai & Hong (2005) presented an investigation of a support vector machine model with genetic algorithms (SVMGs) to accurately forecast arrivals in Barbados. Fernandon (2005) developed a comparison of a neuro-fuzzy model with other quantitative methods for tourism forecasting in Japan. Kon & Turner (2005) compared the forecasting accuracy of the basic structural method (BSM) and the neural network method to find the best structure for neural network models, using data for arrivals to Singapore.