Research on Global Ship Cargo Capacity Prediction Based on Multi-source Heterogeneous Data

Shuhang Chen^1,2Zhihuan Wang^3,4*Tianye Lu³Jiayang Zhu³CHUNCHANG ZHANG^3,4Xiangming Zeng^3,4Jiayi Wang¹Enmei Tu^1*

• ¹COSCO SHIPPING Technology CO.,Ltd, Shanghai, China
• ²Tongji University, Shanghai, Shanghai Municipality, China
• ³Shanghai Maritime University, pudong, Shanghai, China
• ⁴National Engineering Research Center for Special Equipment and Power Systems of Ships and Marine Engineering, Shanghai, China

Maritime cargo capacity serves as a critical metric for evaluating port operational efficiency and regional economic impact. However, multiple operational and commercial factors influence the accessibility of reliable cargo capacity data within the shipping industry. This investigation leverages maritime big data to conduct a comparative analysis between conventional empirical formulations and machine learning approaches. We integrate multi-source heterogeneous datasets including ship inbound and outbound data, ship archive data, Automatic Identification System (AIS) data. The results demonstrate that the K-nearest neighbors (KNN) algorithm outperforms traditional empirical methods by 19 percentage points in validation set accuracy, achieving 88% predictive precision compared to 69% for conventional approaches. While the model attained 95% accuracy on training data, anomalous vessel operation patterns in validation samples caused the model's predictive accuracy to decrease to 88%, highlighting the need for robust data preprocessing frameworks.