Thailand tourism forecasting based on a hybrid of discrete wavelet decomposition and NARX neural network
| DOI | https://doi.org/10.1108/IMDS-11-2015-0463 |
| Published date | 11 July 2016 |
| Date | 11 July 2016 |
| Pages | 1242-1258 |
| Author | Ratree Kummong,Siriporn Supratid |
| Subject Matter | Information & knowledge management,Information systems,Data management systems |
Thailand tourism forecasting
based on a hybrid of discrete
wavelet decomposition and
NARX neural network
Ratree Kummong and Siriporn Supratid
College of Information and Communication Technology, Rangsit University,
Pathumthani, Thailand
Abstract
Purpose –Accurate forecast of tourist arrivals is crucial for Thailand since the tourism industry is a
major economic factor of the country. However, a nonstationarity, normally consisted in nonlinear
tourism time series can seriously ruin the forecasting computation. The purpose of this paper is to
propose a hybrid forecasting method, namely discrete wavelet decomposition (DWD)-NARX, which
combines DWD and the nonlinear autoregressive neural network with exogenous input (NARX) to
cope with such nonstationarity, as a consequence, improve the effectiveness of the demand-side
management activities.
Design/methodology/approach –According to DWD-NARX, wavelet decomposition is executed
for efficiently extracting the hidden significant, temporal features contained in the nonstationary time
series. Then, each extracted feature set at a particular resolution level along with a relative price as an
exogenous input factor are fed into NARX for further forecasting. Finally, the forecasting results are
reconstructed. Forecasting performance measures rely on mean absolute percentage error, mean
absolute error as well as mean square error. Model overfitting avoidance is also considered.
Findings –The results indicate the superiority of the DWD-NARX over other efficient related
neural forecasters in the cases of high forecasting performance rate as well as competently coping with
model overfitting.
Research limitations/implications –The scope of this study is confined to Thailand tourist
arrivals forecast based on short-term projection. To resolve such limitations, future research should
aim to apply the generalization capability of DWD-NARX on other domains of managerial time series
forecast under long-term projection environment. However, the exogenous input factor is to be
empirically revised on domain-by-domain basis.
Originality/value –Few works have been implemented either to handle the nonstationarity,
consisted in nonlinear, unpredictable time series, or to achieve great success on finding an appropriate
and effective exogenous forecasting input. This study applies DWD to attain efficient feature
extraction; then, utilizes the competent forecaster, NARX. This would comprehensively and specifically
deal with the nonstationarity difficulties at once. In addition, this study finds the effectiveness of
simply using a relative price, generated based on six top-ranked original tourist countries as an
exogenous forecasting input.
Keywords Discrete wavelet decomposition,
Nonlinear autoregressive neural network with exogenous input, Nonstationary time series
Paper type Research paper
1. Introduction
Tourism is vital for several countries, as a result of a large consumption of money
for businesseswith their goods and services as wellas the opportunity for employment in
the service industries, includingairline transportation and accommodation services.As a
consequence, tourism both directly and indirectly contributes a significant share to
nation’s gross domestic product (GDP). Unlike manufacturing, retail trade or
Industrial Management & Data
Systems
Vol. 116 No. 6, 2016
pp. 1242-1258
©Emerald Group Publishing Limited
0263-5577
DOI 10.1108/IMDS-11-2015-0463
Received 12 November 2015
Revised 22 January 2016
Accepted 12 February 2016
The current issue and full text archive of this journal is available on Emerald Insight at:
www.emeraldinsight.com/0263-5577.htm
1242
IMDS
116,6
construction, tourism industrial product and services, e.g., hotel rooms, airline seats and
car rentals have a perishable nature; they cannot be inventoried. Hence, to avoid the
financial cost of excess capacity or the opportunity costs of unfulfilled demand, tourism
management planningis crucial. In this process, the accurate forecastof tourism volume
in the form of arrivals is especially important because it is an indicator of precise future
demand, thereby providing basic information for subsequent managerial planning as
well as decision and policy making. Management sectors, including private as well as
government ones can use such basic information to plan their future operations and to
foresee the need for facilities and infrastructure development.
The purpose of this study is to propose a novel forecasting model, a hybrid between
discrete wavelet decomposition (DWD) (Mallat, 1989) and a nonlinear autoregressive
neural network with exogenous factors (NARX) (Chen et al., 1990; Narendra and
Parthasarathy, 1990), namely DWD-NARX. The proposed method can be used by
hospitality managers as well as forecast practitioners to produce accurate forecasts of
tourist flows to Thailand. NARX, one of proficiently forecasting tools exploits recurrent
neural architecture (Elman, 1990). As opposed to otherrecurrent neural networks (RNN),
it has limited feedback architectures that come only from the output neuron instead of
from hidden neurons. It has been reportedthat such learning architecture canyield more
effective results in NARXmodel than in other recurrent architectures with hidden states
(Horne and Giles, 1995). In order to enhance NARX forecasting performance, a relative
price is employed here as an exogenous input as it economically influences tourists’
destination-decision choice in some significance. However, nonstationarity usually
existed in nonlinear tourism time series possibly severely degrades the forecasting
performance. DWD, therefore is employed here for effectively coping with such
nonstationarity problem. With respect to the proposed method, DWD is executed to
achieve efficient feature extraction for an individual level of frequency resolution in the
tourism timeseries environment. Then, the extractedfeature set at a particular frequency
resolution level along with a relative price as an exogenous input factor are fed into
NARX for further forecasting. At last, the forecasting outputs from all the resolution
levels are reconstructed. An inbound tourism demand to Thailand, which is one of the
main tourist destinations in Asia is focussed in this paper. DWD-NARX is tested with
Thailand tourismmonthly time series over January 1999 to December2013. Performance
evaluation relies on mean absolute percentage errors (MAPE), mean absolute error
(MAE) as well as mean square error (MSE). The remaining parts of the paper are
organizedas follows. In Sections 1.1 and 1.2 motivationfor Thailand tourism forecastand
literature review are respectively mentioned. NARX and wavelet decomposition are
briefly reviewed in Section 2. The proposed DWD-NARX is described in Section 3.
Section 4 delineates experimental comparisons among the proposed method and some
related classical neural forecasters. Managerial implications are stated in Section 5. The
overall conclusion is drawn in the last section.
1.1 Motivation for Thailand tourism forecast
Thailand is worldwide known as the “Land of Smiles.”Time magazine reported in 2013
(Quan, 2013) that Bangkok, the capital of Thailand was identified as one of the most
visited city in the world by the global destination cities index; while Suvarnabhumi
Airport and Siam Paragon Shopping Mall were the world’s most geotagged location on
Instagram. In addition, a famous tourism destination in Thailand is represented by an
elegant, attractive decorated Wat Phra Kaew, also known as the Temple of the Emerald
Buddha or the Grand Palace (HikerBays, n.d.). Moreover, there are not only several
1243
DWD and
NARX neural
network
To continue reading
Request your trial