Inter-country trade similarity graph-based long short-term memory for port throughput prediction

Abstract

In this study, we propose a new port throughput prediction method called inter-country trade similarity graph-based long-short term memory, designed to enhance the efficiency and reliability of port operations. We introduce the inter-country trade similarity graph, a novel representation that not only captures port throughput trends but also integrates trade transaction information between trading countries and the target port, providing a comprehensive perspective for throughput prediction. For graph generation, an inter-country trade similarity is determined by calculating the similarity of trade transaction information between trading countries via text mining techniques. Then, we develop an end-to-end prediction model based on graph-based sequential learning for the inter-country trade similarity graph. By leveraging both the port throughput trends and the inter-country trade relationships encapsulated in the graph, our model delivers precise throughput predictions. We implemented an empirical analysis and model validation in terms of liquid bulk cargo and dry bulk cargo. Experimental results demonstrate that our model outperforms state-of-the-art baseline methods in terms of root mean squared error and unscaled mean bounded relative absolute error, achieving a remarkable accuracy improvement of 6.9% and 13.0% for liquid bulk cargo and 4.7% and 14.5% for dry bulk cargo, highlighting its effectiveness in real-world applications. We believe that our model is a useful tool for strategic decision-making in port-related industries.