Article: Comparative analysis of conventional and artificial intelligence forecasting models for international tourist arrivals in three metropolitan hubs Journal: Journal for International Business and Entrepreneurship Development (JIBED) 2024 Vol.16 No.3 pp.349 - 375 Abstract: The purpose of this study is to forecast the arrivals of international tourists to three major cities: New York, Singapore and Dubai, based on data procured before the COVID-19 pandemic was declared. We apply four distinct forecasting models: two conventional linear models, namely exponential smoothing and SARIMA, and two advanced non-linear models, specifically the Prophet with Fourier transformation and LSTM utilising deep learning techniques. We used monthly arrival data spanning from January 2001 to December 2017. Our findings reveal the superior forecasting performance of LSTM neural networks over dynamic regression with Fourier, resulting in a substantial reduction in error rates ranging from 20% to 60%. Notably, SARIMA outperformed conventional models in certain assessments. Despite their accuracy, these models retain generalisability, which is also a significant advance for practitioners such as policymakers and decision makers. This enhanced forecasting capability empowers decision-makers to plan infrastructure and human resource requirements with increased confidence in future endeavours. Inderscience Publishers - linking academia, business and industry through research

Title: Comparative analysis of conventional and artificial intelligence forecasting models for international tourist arrivals in three metropolitan hubs

Authors: Mohammed Al-Shehhi; Andreas Karathanasopoulos; Mohamed Osman; Ibrahim Tabche

Addresses: Dubai Business School, University of Dubai, UAE ' Dubai Business School, University of Dubai, UAE ' Dubai Business School, University of Dubai, UAE ' School of Management, Canadian University Dubai, Dubai, UAE

Abstract: The purpose of this study is to forecast the arrivals of international tourists to three major cities: New York, Singapore and Dubai, based on data procured before the COVID-19 pandemic was declared. We apply four distinct forecasting models: two conventional linear models, namely exponential smoothing and SARIMA, and two advanced non-linear models, specifically the Prophet with Fourier transformation and LSTM utilising deep learning techniques. We used monthly arrival data spanning from January 2001 to December 2017. Our findings reveal the superior forecasting performance of LSTM neural networks over dynamic regression with Fourier, resulting in a substantial reduction in error rates ranging from 20% to 60%. Notably, SARIMA outperformed conventional models in certain assessments. Despite their accuracy, these models retain generalisability, which is also a significant advance for practitioners such as policymakers and decision makers. This enhanced forecasting capability empowers decision-makers to plan infrastructure and human resource requirements with increased confidence in future endeavours.

Keywords: forecasting tourism arrivals; SARIMA; Holt-Winters TES; dynamic harmonic regression; DHR; long short-term memory; LSTM deep learning; machine learning; ML.

DOI: 10.1504/JIBED.2024.142489

Journal for International Business and Entrepreneurship Development, 2024 Vol.16 No.3, pp.349 - 375

Received: 27 Mar 2024
Accepted: 02 Apr 2024
Published online: 04 Nov 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article

Title: Comparative analysis of conventional and artificial intelligence forecasting models for international tourist arrivals in three metropolitan hubs

Keep up-to-date