Pannee Suanpang, Pitchaya Jamjuntr
This research aims to present a comparison of deep learning techniques when working with time-series data to predict the recovery and resilience of tourism businesses in Thailand. This type of business generates a lot of income and plays a role in the development of the country. However, Thailand's tourism industry although has its identity and strengths, but it cannot survive the impact of the COVID-19 epidemic. The biggest impacts of COVID-19 to tourism business are unemployment of the population and the slowdown in national development. Recognizing the importance of tourism as mentioned above, the researchers compared the functioning of three deep learning methods which consists of Vanilla LSTM, Bidirectional LSTM, and Stacked LSTM, when used with time-series data during the year 2020-2024 to planning the rehabilitation of Thailand's tourism business. The experimental results show that, Bidirectional LSTM is the most effective deep learning model when working with time-series data compared to Vanilla LSTM and Stacked LSTM, by considering average error of the Bidirectional LSTM when working with time-series data, which is equal to 5.85916575 which has a lower average error value less than that of the average error of Vanilla LSTM and Stacked LSTM, which are equal to 5.91683425 and 5.92949325 respectively. The results of this research can be concluded that, the deep learning with Bidirectional LSTM approach is the most efficient choice for working with time-series data, when applied to 5 years data and compare with the Vanilla LSTM and Stacked LSTM. In addition, this research also shows a good practice to implement deep learning to the tourism business.