ANN-polynomial-Fourier series modeling and Monte Carlo forecasting of tourism data

Abstract

Modeling and forecasting of tourism data have received attention in the past decades. Turkey is one of the countries that benefit significantly from the tourism industry. Several time-series models have been recommended to best describe tourist arrivals to Turkey. However, in the 21st century, the world experiences great uncertainty in most possible event outcomes. These uncertainties are very difficult to account for. We proposed a hybrid artificial neural network (ANN)-polynomial-Fourier method to model the number of foreign visitors to Turkey from January 2004 to December 2020. The proposed model performance before and during the COVID-19 pandemic is evaluated separately. We evaluate the model performance by comparing with results from Danbatta and Varol (2021, ), Fourier series, and ARIMA models. To account for prediction uncertainties, we ran 300 Monte Carlo simulations within +/- 2 sigma from the model regression curve. According to the result outcomes, the proposed ANN-polynomial-Fourier has proven worthy to be considered a candidate model for the Turkish tourism data. The multistep ahead forecast suggests a 10.22% increase in the monthly foreign visitors' arrivals to Turkey in the year 2021.