International Journal of Engineering Research and Technology. ISSN 0974-3154, Volume 12, Number 11 (2019), pp. 1864-1868 © International Research Publication House. http://www.irphouse.com 1864 Forecasting Foreign Tourist Arrivals to Bali: Hybrid Double Exponential Smoothing Approach Seng Hansun, Marcel Bonar Kristanda Department of Informatics, Universitas Multimedia Nusantara, Tangerang, Indonesia. Abstract Bali is one of the most popular tourist destinations in the world. With its magnificent nature and enchanting culture, it has attracted many people to come and visit every year. The need to have a proper and accurate tourist flow prediction has become important since it could be a scientific reference for decision making process for the development of tourism sectors in Bali province. In this research, we try to forecast the foreign tourist arrivals to Bali by using its monthly distribution data recorded from January 2008 to December 2018. Two hybrid double exponential smoothing methods, i.e. B-WEMA and H- WEMA, were successfully implemented and could predict the future foreign tourist arrivals number. B-WEMA excels H- WEMA in terms of accuracy level which was calculated using MSE and MAPE error measurements. Furthermore, the prediction of foreign tourist arrivals to Bali province for 2019 was given which showed a slight increase compare to the number of foreign tourist arrivals in 2018. Keywords: Bali, Forecasting, Foreign tourist arrivals, Hybrid double exponential smoothing, monthly distribution data. I. INTRODUCTION One of the key development factors and income sources for many developing countries is the tourism sector. It is the main factor for export income, job creation, and local development [1]. In Indonesia, it has become the third largest GDP contributor after oil and mining sectors [1]. Accurate prediction of tourist flow has become a key issue in tourism economic analysis and development planning [2]. In fact, it has attracted much attention from notable researchers and practitioners, as we can see from the increasing academic literature in this domain [3]. Some of them using statistical approach, as we can see in the works of Hopken et al. [4], Zhu et al. [5], and Tung [6]. Others used artificial intelligence and machine learning approach, such as integrated fuzzy time series model [7] and Long Short-Term Memory (LSTM) network [8]. Some others even proposed hybrid approaches as the works of Liu et al. [9], Sun et al. [10], and Binru et al. [11]. One of the most popular tourist destinations in the world is Bali province [12]. It is located in Indonesia, more exactly at 8°3’40” - 8°50’48” S dan 114°25’53” - 115°42’40” E [13]. Figure 1 shows the Bali island map. Fig. 1. Bali island map [14] With its magnificent nature and enchanting culture, Bali has attracted many people around the world to come [15]. Bali Government Tourism Office has reported an increasing trend of foreign tourist arrivals to Bali each year. In 2016, around 4.9 million foreign tourists had come to Bali and increased to 5.7 million people in 2017, and increased again in 2018 to 6.0 million people [16]. Therefore, there is a need for a proper and accurate foreign tourist flow prediction to Bali province, since it could become a scientific reference for decision making process of tourism related departments. In this research, we try to forecast foreign tourist arrivals to Bali by its monthly distribution data. Two hybrid double exponential smoothing methods will be incorporated here, i.e. the Brown’s Weighted Exponential Moving Average (B- WEMA) and Holt’s Weighted Exponential Moving Average (H-WEMA) methods. Both of them are improved version of original WEMA method which was introduced in 2013 [17]. Moreover, to get the accuracy level of both methods applied in this study, we used Mean Square Error (MSE) and Mean Absolute Percentage Error (MAPE), two of the most common used forecast error measurement criteria in time series analysis. In the next section, we will give more detail of B-WEMA and H-WEMA followed by the explanation of MSE and MAPE in Section 3. The implementation results and analysis will be given in Section 4, and some conclusion remarks will end the organization of this paper.