Application of machine learning in port throughput prediction

The rapid growth of global trade has made ports indispensable core nodes in the global logistics network. The study of port throughput has become a hot topic for modern researchers. This study focuses on Guangzhou Port Group, extracting its throughput data from 2014 to 2023 for preliminary time series analysis. By observing the autocorrelation and partial autocorrelation graphs based on monthly throughput data, we explored how port throughput is influenced by changes in industrial structure, the economic conditions of the hinterland, transportation capacity, and unexpected factors. Furthermore, through regression analysis, we identified key factors significantly impacting port throughput, including total import and export trade volume (in billions of RMB), total retail sales of consumer goods (in ten thousand RMB), waterway freight volume (in million tons), fixed asset investment growth rate (%), and overseas pandemic death tolls. Additionally, this study developed a combined RF-SVM-XGB port throughput prediction model, improving the average error by 3 percentage points. In terms of predictive performance and model error, the combined prediction model demonstrated better fitting accuracy. Finally, to enhance the model’s stability and generalization ability, cross-validation was used for result validation. Feasibility suggestions for port development were also provided.