Elsevier

Annals of Tourism Research

Volume 75, March 2019, Pages 410-423
Annals of Tourism Research

Tourism demand forecasting: A deep learning approach

https://doi.org/10.1016/j.annals.2019.01.014Get rights and content

Highlights

  • A deep learning method is presented to forecast tourist demand.

  • The introduced method represents an automated approach to feature engineering.

  • The method overcomes the linearity limitations of existing lag order detection.

  • The case study on Macau confirms the superior performance of the proposed approach.

  • The introduced method can be applied to different tourism destinations.

Abstract

Traditional tourism demand forecasting models may face challenges when massive amounts of search intensity indices are adopted as tourism demand indicators. Using a deep learning approach, this research studied the framework in forecasting monthly Macau tourist arrival volumes. The empirical results demonstrated that the deep learning approach significantly outperforms support vector regression and artificial neural network models. Moreover, the construction and identification of highly relevant features from the proposed deep network architecture provide practitioners with a means of understanding the relationships between various tourist demand forecasting factors and tourist arrival volumes.

This article also launches the Annals of Tourism Research Curated Collection on Tourism Demand Forecasting, a special selection of research in this field

Introduction

Unoccupied hotel rooms, unsold event tickets and unconsumed food items represent unnecessary costs as well as unrealized revenue, a combination that poses a potential threat to financial sustainability. In short, many tourism and hospitality products cannot be stockpiled for future use, making the need for accurate tourism demand forecasting crucial (Frechtling & Frechtling, 2001). As such, accurate tourism demand forecasts provide valuable aid for strategic, tactical and operational decision making (Lim, 1997; Song & Li, 2008). For example, governments need accurate tourism demand forecasts for informed decision making on issues such as infrastructure development, and accommodation site planning (Chan, Hui, & Yuen, 1999); organizations need the forecasts to make tactical decisions related to tourism promotion brochures, and tourism and hospitality practitioners need accurate forecasts for operational decisions such as staffing and scheduling. Accordingly, accurate tourism demand forecasting is an essential element that provides crucial information for tourism-related decision making.

The majority of tourism demand forecasting studies fall under the well-established category of quantitative approach, which constructs the model from training data on past tourist arrival volumes and various tourism demand forecasting factors (Song & Li, 2008; Wu, Song, & Shen, 2017). With the advances in Web technology, search engines have become essential for tourists in planning their trips by obtaining destination information on hotels, attractions, and weather. SII data have been acknowledged as a potential indicator of tourism demand in the destination market (Dergiades, Mavragani, & Pan, 2018; Fesenmaier, Xiang, Pan, & Law, 2011; Yang, Pan, & Song, 2014), and researchers have examined Search Intensity Indices (SII) data for tourism demand forecasting (Volchek, Liu, Song, & Buhalis, 2018; Xiang & Pan, 2011).

Although incorporating SII data is promising for accurate tourism demand forecasting, some practical challenges have arisen for practitioners attempting to use them with traditional forecasting models. More specifically, the following two practical barriers exist.

The first barrier is related to feature engineering. As mentioned by Song and Li (2008), a large number of factors have been considered as potential tourism demand forecasting determinants or indicators, examples include exchange rate, tourism prices, travel costs and various SII data. As the number of potentially influential factors increases, available training data in the feature space become increasingly sparse. In tourism demand forecasting, this means insufficient data for the construction of a reliable model. Many forecasting models have difficulties in learning from the training data with too many explanatory factors (Guyon & Elisseeff, 2003). Therefore, feature engineering has been an important step in forecasting model construction because it focuses on extracting the best set of relevant features from a large variety of potential factors (Zhang, Zhang, & Yang, 2003).

Even though the meaning of factors, such as search engine keywords, are largely known, in the real world scenario, thousands of potential keywords may be related to a destination tourism market. Currently, feature engineering for tourism demand forecasting depends largely on domain knowledge on the destination tourism market and requires significant human efforts in selecting effective features (Xiang & Pan, 2011; Yang, Pan, Evans, & Lv, 2015).

Lag order selection is the second barrier. Despite an increasing number of tourism demand forecasting methods adopting the SII data, only a small number of studies detect the lead or lag relationships between time series data. Most existing works examined the hypothesis of no predictability through Pearson correlation coefficients or the Granger causality test (Dergiades et al., 2018; Li, Pan, Law, & Huang, 2017), in which the null hypothesis is investigated by testing whether the lagged values of a factor are strongly related or contributing significantly to tourist arrival volume. However, neither Pearson correlation coefficients or Granger causality test works reliably when the underlying relationship is nonlinear (Reshef et al., 2011). Hence, the capability of selecting all potentially interesting relationships in a dataset will allow tremendous versatility in the construction of more accurate forecasting models.

Time series, econometric and artificial intelligence models provide excellent forecasting performance, and they break the barrier of feature engineering based on the domain knowledge of the destination market. However, adopting existing forecasting methods for every destination market is inconvenient because of the effect from complicated real-world situations. This is especially true when massive amounts of SII data are adopted as tourism demand indicators during which they may require significant domain expertise to confine the uncertainty. Moreover, for each SII feature, the number of effective lags may be different as well. The difficulty increases when combined with other issues, such as the language bias and platform bias (Dergiades et al., 2018).

Recent advances in artificial intelligence, especially the deep learning techniques, have provided methods of breaking the above barriers and enabling more accurate tourism demand forecasting (Pouyanfar et al., 2018). Deep network architectures extend the artificial neural network models with more than two nonlinear processing layers, and have been proven effective for various applications. Their success is attributed usually to their built-in feature engineering capability, which motivates us to break those two barriers simultaneously within the machine learning process. With regard to the contextual information for time series analysis, deep network architectures also have certain advantages in flexible yet discriminative non-linear relationships. Specifically, Recurrent Neural Network (RNN), Long-Short-Term-Memory (LSTM) and Attention Mechanism are capable of handling and learning long-term dependencies. These properties make deep learning an alternative solution to tourism demand forecasting. In this paper, we aim to fill the void by proposing a deep learning approach to tourism demand forecasting and address the previously mentioned two practical barriers simultaneously.

The rest of this paper is organized as follows. The Literature review section reviews the related literature on tourism demand forecasting, and introduces the deep learning technique. The Methodology section describes the conceptual framework of tourism demand forecasting with deep learning. The Empirical study section provides a case study on Macau tourist arrivals, and analyzes the comparison results with baseline methods. Finally, conclusions and implications for future works are summarized in the Conclusions section.

Section snippets

Literature review

Tourism is an important source of economic growth as well as foreign exchange earnings and jobs creation. Accurate tourism demand forecasts are paramount because they provide vital information to tourism practitioners and researchers for making decisions on activities such as earmarking resources, as well as identifying priorities and potential risks. This section provides a brief literature review on tourism demand forecasting studies and the deep learning technique that motivates this work.

Methodology

This study proposes a conceptual framework of tourism demand forecasting with deep learning, and then describes the deep network architecture that addresses the above mentioned challenges.

Empirical study

To empirically investigate the prediction performance of the proposed conceptual framework, we conduct an empirical study on the forecasting of monthly tourist arrival volume in Macau. Macau is an autonomous region of China, and it is across the Pearl River Delta from Hong Kong. Gaming and tourism make up the pillar industry of Macau, and they contribute remarkably to the economic growth of the city. Thus, maintaining a timely and accurate forecasting of tourist arrival volumes is essential to

Conclusions

In the tourism industry, precise and timely demand forecasts are critical for informed decision making by most, if not all, providers of products and services. Time-series, econometrics and AI models have been extensively examined in the past decades. Traditionally, the accuracy of tourism demand forecasting models rely on the goodness of the set of features. Poor selection of determinants or indicators often leads to lower accuracy compared with that of good selection. The selection of

Acknowledgement

The work described in this paper was partially supported by the National Natural Science Foundation of China under Grant No. 71471011, International Cooperation Project of Institute of Information Engineering, Chinese Academy of Sciences under Grant No. Y7Z0511101, and research grants funded by the Hong Kong Polytechnic University (G-YBXG, G-UAE8).

Rob Law, [email protected], Ph.D. His research interests are technology applications to tourism and information management.

References (61)

  • R. Law

    Back-propagation learning in improving the accuracy of neural network-based tourism demand forecasting

    Tourism Management

    (2000)
  • R. Law et al.

    A neural network model to forecast Japanese demand for travel to Hong Kong

    Tourism Management

    (1999)
  • X. Li et al.

    Forecasting tourism demand with composite search index

    Tourism Management

    (2017)
  • C. Lim

    Review of international tourism demand models

    Annals of Tourism Research

    (1997)
  • L.F. Martins et al.

    An empirical analysis of the influence of macroeconomic determinants on World tourism demand

    Tourism Management

    (2017)
  • M. Meleddu et al.

    Evaluation of individuals' intention to pay a premium price for ecotourism: An exploratory study

    Journal of Behavioral and Experimental Economics

    (2016)
  • R. Rivera

    A dynamic linear model to forecast hotel registrations in Puerto Rico using Google Trends data

    Tourism Management

    (2016)
  • S. Shen et al.

    Combination forecasts of International tourism demand

    Annals of Tourism Research

    (2011)
  • H. Song et al.

    Tourism demand modelling and forecasting. A review of recent research

    Tourism Management

    (2008)
  • H. Song et al.

    Forecasting international tourist flows to Macau

    Tourism Management

    (2006)
  • S.F. Witt et al.

    Forecasting tourism demand: A review of empirical research

    International Journal of Forecasting

    (1995)
  • Z. Xiang et al.

    Travel queries on cities in the United States: Implications for search engine marketing for tourist destinations

    Tourism Management

    (2011)
  • X. Yang et al.

    Forecasting Chinese tourist volume with search engine data

    Tourism Management

    (2015)
  • M. Alvarez-Diaz et al.

    Forecasting tourist arrivals to Balearic Islands using genetic programming

    International Journal of Computational Economics and Econometrics

    (2009)
  • W.P. Andrew et al.

    Forecasting hotel occupancy rates with time series models: An empirical analysis

    Hospitality Research Journal

    (1990)
  • P. Baldi et al.

    The Principled design of large-scale recursive neural network architectures-DAG-RNNs and the protein structure prediction problem

    Journal of Machine Learning Research

    (2003)
  • W. Bao et al.

    A deep learning framework for financial time series using stacked autoencoders and long-short term memory

    PLoS One

    (2017)
  • G.E. Box et al.

    Time series analysis: Forecasting and control

    (2015)
  • Y.-M. Chan et al.

    Modeling the impact of sudden environmental changes on visitor arrival forecasts: The case of the Gulf War

    Journal of Travel Research

    (1999)
  • K. Cho et al.

    Learning phrase representations using RNN encoder-decoder for statistical machine translation

  • Cited by (265)

    View all citing articles on Scopus

    Rob Law, [email protected], Ph.D. His research interests are technology applications to tourism and information management.

    Gang Li, [email protected], Ph.D. His research interests are technology applications to tourism and hospitality, data science, and business intelligence.

    Davis Ka Chio Fong, [email protected], Ph.D. His research interests are social and economic impacts of gambling, responsible gambling, and tourist behaviour.

    Xin Han, [email protected], Master by research student, his research interests are tourism and hospitality management, and data science.

    View full text