Long-Short Term Memory (LSTM) Based Architecture for Forecasting Tourist Arrivals

Abstract

Indonesia is a country that has beautiful destinations. Indonesia’s natural beauty can increase the arrival of foreign tourists if appropriately managed. The arrival of foreign tourists will be able to improve the economy of the community around the destination. But in reality foreign tourist arrivals sometimes go up and sometimes down. This can be seen in the data obtained from the Indonesian Central Bureau of Statistics (BPS) for 2006–2021.The purpose of this study is to forecast tourist arrivals to Indonesia using the Long Short Term Memory (LSTM) method because the method is capable of handling sequential data such as tourist arrival data. This study also compares the 2006–2009 dataset divided by 0 and without 0, and the 2021 dataset with 0 and without 0. To get the best results, these results are evaluated using the Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Mean Absolute Error (MAE). The test results obtained for 2006–2019 data with 0 have the best value on MAPE which is 6.51. While the results of the RMSE test with the best value are found in the 2006–2021 data without 0, namely 101618.80 and for testing using MAE the best value is found in the 2006–2021 data, namely with a value of 91922.33. Based on these results, it can be concluded that the more data the more accurate it is if the data is clean or does not have a value of 0.