Publication

Abstract : Efficient ship inspections are vital for minimizing vessel detentions in the maritime industry. This study proposes a hybrid two-phase machine learning approach to assess ship risk levels based on historical detention data and vessel characteristics, compiled from major Memoranda of Understanding (MOUs) such as the Paris and Tokyo MOUs. In Phase 1, a parallel random forest model predicts specific detention code categories. Phase 2 uses an XGBoost classifier to assign an overall risk category to categorize ships into low, medium, or high-risk tiers. This approach combines the strengths of both algorithms for comprehensive risk assessment. The models demonstrated excellent discrimination capability, with a perfect 100% accuracy for the high risk classifier. Our models demonstrate excellent discrimination capability across diverse risk levels with AUC values of 0.99 for the medium risk classifier and a perfect 1.00 for the high risk classifier. Additionally, we utilize explainable AI techniques, such as SHAP analysis, to elucidate the factors influencing the model’s decisions, enhancing transparency and reliability. This study pioneers the Classification of ship risk based on historical detention records, offering a new tool for improving maritime safety and compliance. The integrated methodology aims to enhance predictive accuracy and support smarter, data-driven inspection strategies.