Value of information for vessel turnaround time predictions

Accurate prediction of container vessel turnaround time (VTT) is crucial for berth allocation, resource planning, and overall terminal efficiency. Early forecast horizons (hours to days before arrival) frequently lack reliable timing and move-count data, which reduces model accuracy. We introduce a forward-looking value-of-information (VoI) approach that quantifies how incremental improvements in data availability and quality affect VTT forecasts. Using terminal-internal data from three terminals and two vessel classes, we evaluate seven features by incrementally replacing estimated pre-arrival values with their post-arrival counterparts (5%–100%) across prediction lead times of 1–120 h and assess the resulting performance across three model classes: XGBoost, Random Forest, and Linear Regression. Across models, the main VoI patterns remain consistent. Timing data (ETA/ETD) primarily improves short-horizon accuracy, while planned-move records yield the largest gains when timing information is sparse. The results indicate comparable data-gap thresholds across settings: when baseline availability is low, recovering missing move-related information tends to offer the highest returns, whereas with higher baseline availability, further improvements in timing-related information quality become more beneficial. Accordingly, operators should prioritize closing major data-availability gaps first, especially for move counts, and then invest in improving data accuracy. The VoI approach supports this prioritization by quantifying expected returns, guiding data-driven decisions, and informing incentives for stakeholders to improve terminal forecasting performance.