Predicting Port Congestion at Busan Port Using Machine Learning Algorithms and Temporal Variables

Ports are vital infrastructures for global logistics and trade, with South Korea handling approximately 99.7^% of its import and export cargo via maritime routes. Busan Port, the largest in the country and the seventh-largest container port worldwide, plays a crucial role in both national logistics and regional economic growth. However, congestion in port operations leads to delays, increased costs, and reduced efficiency. This study employs actual data from Busan Port (February to September 2024) to analyze the impact of temporal variables-such as day of the week, time of day, and monthly factors-on port congestion. Machine learning algorithms, including Random Forest, XGBoost, and LightGBM, were utilized to predict congestion levels using a refined dataset of approximately 1.4 million samples. Results indicate that Extra Trees and CatBoost classifiers achieved high accuracy (0.9654) and AUC (0.9952), while Extreme Gradient Boosting reached an AUC of 0.9989, demonstrating exceptional performance. These findings highlight the effectiveness of machine learning in capturing the nonlinear dynamics of port operations. Future research will aim to enhance model performance by expanding data collection and integrating external factors such as weather conditions and cargo characteristics. This study underscores the potential of machine learning tools in optimizing port management and improving logistical efficiency.