Data Pre-Processing for Neural Network-Based Forecasting: Does It Really Matter?

Abstract

Abstract.This study aims to analyze the effects of data pre-processing on the forecasting performance of neural network models. We use three different Artificial Neural Networks techniques to predict tourist demand: multi-layer perceptron, radial basis function and Elman neural networks. The structure of the networks is based on a multiple-output approach. We use official statistical data of inbound international tourism demand to Catalonia (Spain) and compare the forecasting accuracy of four processing methods for the input vector of the networks: levels, growth rates, seasonally adjusted levels and seasonally adjusted growth rates. When comparing the forecasting accuracy of the different inputs for each visitor market and for different forecasting horizons, we obtain significantly better forecasts with levels than with growth rates. We also find that seasonally adjusted series significantly improve the forecasting performance of the networks, which hints at the significance of deseasonalizing the time series when using neural networks with forecasting purposes. These results reveal that, when using seasonal data, neural networks performance can be significantly improved by working directly with seasonally adjusted levels.Keywords: artificial neural networks, forecasting, multiple-output, seasonality, detrending, tourism demand, multilayer perceptron, radial basis function, Elman JEL classification: L83, C53, C45, R11 IntroductionInternational tourism has become one of today's fastest growing industries. Tourism accounts for almost 10% of total international trade and plays a fundamental role in the long-run economic development of many regions (Akkemik 2012;Sigala, Chalkiti 2014). To achieve a sustainable tourism model, policy makers and professionals need more accurate predictions of the number of tourist arrivals at the destination level. Many authors have acknowledged the importance of applying new approaches to tourism demand forecasting in order to improve the accuracy of the methods of analysis (Song, Li 2008). The availability of more advanced forecasting techniques has led to a growing interest Artificial Intelligence (AI) models (Yu, Schwartz 2006;Goh et al. 2008;Lin et al. 2011;Chen 2011;Celotto et al. 2012;Wu et al. 2012;Cang, Yu 2014) to the detriment of time series models (Chu 2008(Chu , 2011Assaf et al. 2011) and causal econometric models (Page et al. 2012). Some of the new AI based techniques are fuzzy time series models (Tsaur, Kuo 2011), genetic algorithms (Hadavandi et al. 2011), expert systems (Shahrabi et al. 2013;Pai et al. 2014) and Support Vector Machines (SVMs) (Chen, Wang 2007;Hong et al. 2011). Recent research has shown the suitability of Artificial Neural Networks (ANNs) for dealing with tourism demand forecasting (Teixeira, Fernandes 2012;Claveria, Torra 2014).In spite of the successful use of ANNs for time series forecasting, very few studies compare the accuracy of different NN architectures for tourism demand forecasting at a regional level. The present study deals with tourist a...

Regional Tourism Demand Forecasting with Machine Learning Models: Gaussian Process Regression vs. Neural Network Models in a Multiple-Input Multiple-Output Setting

“…This result coincides with that of Claveria et al (2017) for Catalonia, who analysed the effects of data pre-processing on the forecasting performance of NN models and found that the predictive accuracy of the models improved with seasonal adjusted data. Medeiros et al (2008) developed an alternative approach to analyse the demand for international tourism in the Balearic Islands.…”

Forecasting tourism demand with denoised neural networks

“…In fact, this study showed that NN struggle to model and forecast when presented with raw data which has strong seasonality and trends. More recent evidence by Claveria et al (2017) provides further support for deseasonalizing series when forecasting with NN. Accordingly, it is clear that in contrast to our personal beliefs, many researchers argue in favour of deseasonalisation or detrending as the more suitable approach for improving the accuracy of NN forecasts.…”

Recurrent Neural Networks for Time Series Forecasting: Current Status and Future Directions

“…However, more recent experiments suggest that deseasonalization prior to feeding data to the NNs is essential since NNs are weak in modelling seasonality. Particularly, Claveria et al (2017) empirically show for a tourism demand forecasting problem that seasonally adjusted data can boost the performance of NNs especially in the case of long forecasting horizons. Zhang and Qi (2005) conclude that using both detrending and deseasonalization can improve the forecasting accuracy of NNs.…”