Research on global ship cargo capacity prediction based on multi-source heterogeneous data

Abstract

Maritime cargo capacity serves as a critical indicator of port efficiency and regional economic impact, yet reliable data remain constrained by operational and commercial complexities. This study addresses this gap by leveraging maritime big data to compare traditional empirical methods with machine learning approaches, integrating multi-source datasets (ship inbound/outbound records, vessel archives, and AIS data). Results demonstrate that the K-nearest neighbors (KNN) algorithm achieves 88% predictive accuracy on validation data—a 19-percentage-point improvement over conventional methods (69%). While training accuracy reached 95%, anomalous vessel operations in validation samples reduced performance to 88%, revealing the model’s sensitivity to real-world variability and underscoring the need for enhanced data preprocessing. These findings highlight machine learning’s potential to refine cargo capacity estimation while emphasizing the importance of robust data quality frameworks for operational deployment.