Time Series Analysis for Predicting Import and Export Volumes Based on Deep Learning

การวิเคราะห์อนุกรมเวลาสำหรับการทำนายปริมาณการนำเข้าและส่งออกสินค้าโดยใช้การเรียนรู้เชิงลึก

Abstract

           A trade deficit often indicates that a country is consuming more goods than it can produce itself. A trade deficit can have both positive and negative impacts on the economy of a country. Therefore, forecasting the overall value of imports and exports of a country in advance is very beneficial for strategic planning to manage potential problems arising from a trade deficit. For these reasons, this research studies, analyzes, and compares time series models that can be used to predict the overall volume of imports and exports using statistical methods and machine learning methods including DNN, LSTM, ARIMA, and Prophet. The Mean Squared Error (MSE) values are 0.4116615060290905, 0.9943514149915439, 0.5022269255490994, and 1.4212623390468428, respectively, while the Root Mean Squared Error (RMSE) values are 0.6416085302028727, 0.9971717078776071, 0.7086797058961822, and 1.1921670768171897, respectively. The experimental results conclude that the DNN model provides the highest efficiency in forecasting both the import and export volumes of goods and services for the country.