Comparison of Deep Neural Network Architectural Models for Predicting Tourist Visits to Bali during the Pandemic Period

Abstract

Tourism plays a critical role in any economy as it provides a source of income for communities. Bali, one of Indonesia's provinces, holds significant potential in the tourism industry, with a majority of its population employed in this sector. However, fluctuations in tourist visits can pose challenges when creating policies to address issues in the field. Therefore, forecasting is necessary to anticipate post-pandemic tourist arrival patterns to ensure a smooth tourism recovery process. Forecasting is a vital tool that assists in making sound decisions. In this study, we utilized three forecasting methods: Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). We took a comparative approach, using these three deep neural network architectures to predict tourist visits to Bali during the pandemic. We tested the architectural models using datasets from Badan Pusat Statistik (BPS) and evaluated the model's performance using RMSE and MAE. The results showed that the LSTM model outperformed the CNN and GRU models, with an RMSE value of 0,329036 and MAE value of 0,285874. Based on the study, we can conclude that the LSTM model performed better and can predict tourist arrivals in Bali with reasonable accuracy