Freight Truck Turnaround Time Prediction at Container Ports Using Transfer Learning

In South Korea, 99.7% of international freight is transported through ports. At ports handling massive cargo volumes, prolonged truck waiting times have become a significant social concern. To enhance port operational efficiency and ensure driver safety, systematic congestion management is required, which can be facilitated by predicting truck turnaround time (TAT) in advance. However, existing TAT prediction studies have focused on individual ports where data collection is feasible, limiting the applicability of these models to other ports. The objective of this study was to evaluate the transferability of TAT prediction models to different ports. For the analysis, digital tachograph data capturing the trajectories of heavy-duty trucks were employed. The results indicate that a long short-term memory-based model effectively captures the complex operational characteristics of ports and demonstrates high predictive accuracy at Busan New Port and Busan North Port. By applying transfer learning from the best-performing Busan New Port model, the predictive accuracy for Gunsan Port, a target port with limited data, was substantially improved. This study confirms the feasibility of applying transfer learning in ports with constrained data availability, demonstrating that practical TAT prediction models can be developed under realistic operational constraints.