Improving anchorage occupancy forecasting with stacked ensemble learning

Anchorage areas are essential for safe and efficient maritime operations. However, conventional forecasting models often underperform in dynamic port conditions, as they rely heavily on historical averages and static assumptions. To address these limitations, this study proposes a forecasting framework for anchorage occupancy. This framework uses stacked ensemble learning, integrating both statistical and machine learning models to enhance predictive accuracy and operational reliability. The proposed approach was applied to occupancy data from the E1 anchorage at Ulsan Port, with performance evaluated across various forecasting models and ensemble strategies. In addition, a hexagon-based occupancy estimation method was implemented to assess spatial efficiency and safety in comparison to the traditional anchor circle method. The results demonstrate that the stacking ensemble model effectively captures complex, nonlinear patterns in vessel traffic and delivers improved forecasting performance. These findings highlight the practical potential of stacking ensemble techniques and spatial modeling innovations in enabling proactive anchorage management, reducing congestion, and enhancing maritime safety in real-world port environments.