Akademik Veri Yönetim Sistemi
RELIABILITY ENGINEERING & SYSTEM SAFETY, cilt.267, 2026 (SCI-Expanded, Scopus)
Effective monitoring of ship compliance with international maritime regulations is vital for ensuring maritime safety and environmental protection. This study introduces a hybrid predictive framework aimed at assessing the likelihood of ship detentions based on Port State Control (PSC) inspection outcomes. The framework integrates four complementary methods: Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance, Random Forest (RF) to identify key predictors, Association Rule Mining (ARM) to uncover underlying patterns, and Logistic Regression (LR) to estimate marginal effects and interpret results. Using inspection data from oil/ chemical tankers under the Paris Memorandum of Understanding (MoU) between April 2021 and April 2024, the study places particular emphasis on detentions associated with deficiencies related to the Ballast Water Management (BWM) Convention. Key variables include ship size, inspection region, deficiency types and numbers, and the safety performance of operating companies. The results reveal that detention risk is significantly influenced by ship characteristics, regional inspection patterns, and both the type and numbers of deficiencies. Notably, BWM-related documentation issues (e.g., management plans, record books), low company performance, and inadequate crew training are strong predictors of detention. Methodologically, the RF model achieved a high predictive accuracy of 84.7 %, while ARM identified association rules with confidence levels up to 98 %, particularly for ships subject to expanded inspections with more than seven deficiencies. LR further confirmed the statistical significance of poor company performance and critical defect types through high odds ratios. Overall, the proposed framework offers a comprehensive, data-driven tool to support PSC authorities in identifying high-risk ships, enabling more efficient and targeted inspections while enhancing regulatory enforcement.