Forecasting link disappearance of global liner shipping network using motif structures and explainable machine learning method

Liner shipping carries about 70% of world’s seaborne trade value. As such, accurately predicting link disappearance between ports in the evolving global liner shipping network (GLSN), which signifies the cessation of direct liner shipping service routes between ports, is critical for mitigating maritime logistics disruptions. This paper develops an explainable machine learning model that leverages network motifs to forecast link disappearance in the yearly snapshots of GLSN. Our proposed model outperforms classical baselines in overall predictive performance. We then deliver performance evaluation at the individual port level, revealing two key findings: link disappearance prediction is more challenging for high-degree ports, yet our proposed model achieves superior accuracy for these high-degree ports as compared with other methods. Finally, an analysis of model interpretation confirms that the mechanisms governing link disappearance differ substantially between high- and low-degree ports. These findings offer actionable insights for port operators to assess connectivity risks of their ports and thus improve strategic decision-making.