“Multi-model tourist forecasting: case study of Kurdistan Region of Iraq”

The tourism industry has been one of the leading service industries in the global econ- omy in recent years and the number of international tourism in 2018 reached 1.4 bil-lion. The goal of the research is to evaluate the performance of various methods for forecasting tourism data and predict the number of tourists during 2019 and 2022. Performance of 15 prediction models (i.e. Local linear structural, Naïve, Holt, Random walk, ARIMA) was compared. Based on error measurements matrix (i.e. RMSE, MAE, MAPE, MASE), the most accurate method was selected to forecast the total number of tourists from 2019 to 2022 to Kurdistan Region (KR), then forecasts were performed for each governorate in KR. The results show that among 15 examined models of tour- ist forecasting in KR, Local linear structural and ARIMA (7,3,0) model performed best. The number of tourists to KR and each governorate in KR is predicted to increase by most experimented models, especially those which demonstrated higher accuracy. Generally, the number of tourist to KR predicted by ARIMA (7,3,0) is a lot bigger than Local linear structure. Linear structural predicted the number increase to 3,137,618 and 3,462,348 in 2020 and 2022, respectively, while ARIMA (7,3,0) predicted the num- ber of tourists to KR to increase rapidly to 3,748,416 and 8,681,398 in 2020 and 2022. Our results demonstrate that among 15 examined models to tourist forecasting in KR, Local linear structural and ARIMA (7,3,0) model best performed based on error measurement metrics such as RMSE, MAE and MASE. The number of tourists to KR and each governorate in KR is predicted to increase by most experimented models, especially those which demonstrated higher accuracy. the of KR we have to consider the effects of different political and factors which affect the number of tourists to KR such level of large number of tourists to Kurdistan.


INTRODUCTION
The tourism sector is of great importance in the economic balance of a large number of countries in the world that have the characteristics and qualifications of tourism. Tourism has continued to develop as a human activity, achieving many advantages, which led many countries to pay attention to them and increase their revenues. The tourism industry has played an important role in achieving the economic development of the countries. Tourism depends on the human factor in a great way. It aims at achieving many occupation opportunities according to the reports of the World Tourism and Travel Council. Tourism has proved to be a strong sector of economic activity and a major contributor to the state economic recovery through generating billions of dollars, on the one hand, and creating occupation opportunities for the region's population, on the other hand (Petrevska, 2017). The tourism industry has been one of the leading service industries in the global economy in recent years through economic flows resulting from tourism (Ekanayake & Long, 2012;USAID, 2008) For example, the number of international tourism in 2018 reached 1.4 billion, an increase of 6% (Chhorn & Chaiboonsri, 2017; UNWTO, 2019). The prediction of the number of tourists to the Kurdistan Region of Iraq for the next period assists in order to the required services and tourism facilities (Unakıtan & Akdemir, 2015).

LITERATURE REVIEW
Being one of the significant fields in tourism studies, tourism request modeling and predicting has involved much consideration from academics and experts (Chhorn & Chaiboonsri, 2017;Song & Li, 2008). There are many models were utilized to forecasting of tourism demands such as ML, linear and nonlinear models, multivariate exponential smoothing model, a MLP ANN, ANN-GARCH model and others. The field of prediction has undergone profound changes. A combination of linear and nonlinear models provides solutions in which models are optimally combined and can be applied to actual situations, for example, predicting economic time series, travel request and exchange rates. ML is founded on the creation of an empirical learning algorithm (C.-C. Lin, C.-L. Lin, & Shyu, 2014). The key ML prediction approaches are support vector regression (SVR) and artificial neural network (ANN) models. Artificial neural network models are used to predict economic development and predict exchange rates through several artificial neural networks. Although for economic modeling and prediction, SVR and ANN models have been used, other ML methods (Gaussian Process Regression (GPR)) are almost unsuitable for prediction purposes (Claveria, Monte, & Torra, 2016). More detailed prediction studies and research reviews were given to Song and Li (2008 (Loganathan & Yahaya, 2010). The research predicts that Malaysia would perceive significant tourism for the next years. For forecasting tourist inflow in Bhutan using Seasonal ARIMA (SARIMA), Singh used the model of (0,1,1) (1,1,1) (Singh, 2013). Ellis used four models, namely, Naïve, Etas, ARMA, THETA, for monthly and yearly data tourism study (Ellis, 2016).
In the Kurdistan Region of Iraq, a number of tourist sites were planned to support local economies and promote employment and growth. In recent years, the number of tourists to tourist areas has increased due to the availability of spare time and holidays for the population. There are a number of aspects that contributed to reviving the tourism sector in the Kurdistan Region of Iraq. The development of the tourism industry has established its position as an attractive destination in the region and secondly the development of strategy and policies in a timely manner. The government has constructed two international airports that are capable of operating direct flights to and from Kurdistan (Altaee, Tofiq, & Jamel, 2017). In addition, there are many natural attractions in the study area (topography, climate, water resources and forests, all of these factors have attracted many tourists to tourist site (USAID, 2008). The period of growth between 2007 and 2013 demonstrates a number of problems that interface in the strategy of the Kurdistan Regional Government, in addition to the ongoing budget crisis with the federal government and the threat to the security of the province by the preacher. These problems include failure in investment and planning of basic infrastructure services (health, transportation, hotels and restaurants). These challenges and problems will have to be addressed in order to flourish tourism as a cornerstone of a structured and diversified economy of the state (Rasaiah, 2016;Cura, Singh, & Talaat, 2017). The objects of current study are evaluating the performances of various models (i.e. Local linear structural, Naïve, Holt, Random walk, ARIMA) for forecasting tourism data and predicting the number of tourists during 2020 and 2022.

Methods
The data of tourist number to Kurdistan Region (KR) from 2007 to 2018, entered in R program. Before forecasting the number of tourists, accuracy assessment of the 15 prediction models (i.e. Naïve, Holt, Random walk, ARIMA) was performed. For evaluating forecast accuracy of methods, the data from 2007 to 2014 were utilized as a training data set, and the data from 2015 to 2018 used as an experiment data set of accuracy measures. Then, based on error measurements matrix (i.e. RMSE, MAE, MAPE, MASE), the most accurate method was selected to forecast the number total of tourists during 2019 to 2022 to KR, then, forecasts were performed for each governorate in KR, especially during 2020 and 2022.

Forecasting models
In order to select more suitable forecasting methods, in this study, 15 prediction models were compared.

Accuracy assessment measurements
In order to compare the accuracy of forecasting methods, in this study, the following accuracy assessment matrix was utilized.
1. Mean Error (ME) is calculated according to:  However, the majority of models predicted the number of tourists to KR will increase in 4 coming years and the number predicted to be between 3,000,000 and 3,700,000 in 2022 accord-ing to Naïve, Random walk with drift, Simple exponential smoothing, Holt, HoltWinters and Theta ( Figure 1).
Among examined models to tourist forecasting, especially those well performed for demonstration the trend of the forecast, Local linear structural and ARIMA (7,3,0) model best performed based on error measurement metrics like RMSE, MAE and MASE. Therefore, they selected to forecast the number of tourists during 2020 and 2022 in KR.
Generally, predicted numbers of tourist to KR in ARIMA (7,3,0) is very bigger than Local linear structure. In Table 2, linear structural predicted the number to be 3,137,618, with probability of it to be 4,436,332 in 80% high and 1,838,903 in 80% low interval. In 2022, the number was predicted to increase to 3,462,348 with the probability of it to be 5,404,043 in 80% low interval and 1,520,654 in 80% low interval. Table 3 demonstrates that ARIMA (7,3,0) predicted the number of tourists to KR to increase to 3,748,416 in 2020 and will increase dramatically in 2022 to 8,681,398. The number of tourists to Erbil Governorate according to Local linear structural model forecast be 1,693,656 with probability to decrease to 806,581 in 80% low interval and increase to 2,580,730 in 80% high interval ( Figure 2O and Table 4). In 2022, the number predicted to be 1,874,841 with possibility to be between 548,586 and 3,201,097 in 80% interval. ARIMA (7,3,0) model suggested the number of tourists to Erbil Governorate is very higher than the suggested number in the linear structural model. It predicted to be 1,067,482 in 2020 and in 2022 it increases dramatically to 3,129,778 ( Figure 2L and Table 5).     in 80% high interval ( Figure 3O and Table 5).
In 2022, the number predicted to be 1,281,111. ARIMA (10,1,10) model suggested the number of tourists to Sulaymaniyah Governorate is very higher than the suggested number in the linear structural model. It predicted to be 1,310,090 and 1,591,304 in 2020 and in 2022, respectively ( Figure 3L).

Tourist forecasting of Duhok Governorate in 2020 and 2022
Among examined models to tourist forecasting, Local linear structural and ARIMA (7,3,1) model best performed based on error measurement metrics. The number of tourists to Duhok Governorate in 2020 according to Local linear structural model forecast be 297,698 with probability to decrease to 64,857 in 80% low interval and increase to 530,538 in 80% high interval ( Figure 4O and Table 7). In 2022 the number of tourists predicted to increase slightly to 319,137. According to ARIMA (7,3,1) model, it predicted to increase to 656,791 and 580,708 in 2020 and in 2022, respectively ( Figure 4L).   Our results demonstrate that among 15 examined models to tourist forecasting in KR, Local linear structural and ARIMA (7,3,0) model best performed based on error measurement metrics such as RMSE, MAE and MASE. The number of tourists to KR and each governorate in KR is predicted to increase by most experimented models, especially those which demonstrated higher accuracy.
Generally, predicted numbers of tourist to KR in ARIMA (7,3,0) is very bigger than Local linear structure. Linear structural predicted the number increase to 3,137,618 and 3,462,348 in 2020 and 2022, respectively, while ARIMA (7,3,0) predicted the number of tourists to KR increase rapidly to 3,748,416 and 8,681,398 in 2020 and 2022. However, we have to consider the effects of different political and economic factors which affect the number of tourists to KR such as war, security, level of GDP.
Increasing the number of tourists in the near future (four coming years) requires the KR government to prepare a plan to improve tourism facilities of both general and private sector such as hotels, motels, and roads. In addition, the increase of security and safety in the area is the main factor to attract a large number of tourists to Kurdistan.