A Data-Driven framework for predicting ship berthing time and optimizing port operations at Tan Cang Cat Lai Port, Vietnam

Abstract

This study explores the use of machine learning models to predict ship berthing time (SBT) at Tan Cang Cat Lai port (VNSGN) in Vietnam, identifying Random Forest as the most effective approach. Comparative evaluations of models, including LSBoost, Decision Trees, ANN, CART, Ridge Regression, and Elastic Net Regression, highlight Random Forest’s superior performance with the lowest MAE (1.33), MSE (3.66), and RMSE (1.91), along with the highest R2 (0.84) and accuracy (69.17%). Feature importance analysis reveals cargo volume (35.3%), deadweight, and weather conditions as critical factors influencing SBT. The model demonstrates a strong ability to predict high-demand periods when berth congestion and cargo inflow are highest. By leveraging these insights, ports can optimize berth scheduling, resource allocation, and operational workflows, reducing vessel idle times and improving efficiency. Additionally, the use of predictive tools like Random Forest (RF) supports sustainable logistics by minimizing fuel consumption and greenhouse gas emissions. This research highlights the potential of data-driven approaches in maritime logistics, offering actionable strategies to enhance operational performance, reduce costs, and achieve sustainable port development. The findings provide a foundation for further advancements in predictive modeling for port operations.