Enhancing ship collision risk assessment by integrating virtual ship-based shared nearest neighbor clustering and game-theoretic modeling

Abstract

With the rapid growth in global shipping activities, the assessment of vessel collision risk has become a critical concern for maritime safety management. This study develops a comprehensive framework for identifying high-risk areas in congested waterways by integrating the Shared Nearest Neighbor Density-Based Spatial Clustering of Applications with Noise (SNN-DBSCAN) algorithm and Stackelberg game-theoretic model. The proposed framework first applies SNN-DBSCAN to enable robust waterway regionalization and vessel clustering under highly heterogeneous traffic densities, enhancing the accuracy and efficiency of collision risk assessment. To prevent critical crossing interactions from being fragmented by purely spatial clustering, we introduce a virtual-ship representation based on a risk-invariance principle, ensuring that high-risk encounters are preserved in the interaction set. Furthermore, a leader-follower game is employed to characterize strategic vessel responses by jointly considering safety, efficiency, and decision uncertainty to predict the next actions of the target vessel. The proposed framework is validated using empirical data from the busy waters of Hong Kong under daytime, nighttime, heavy precipitation, and strong winds. The results reveal pronounced scenario-dependent changes in vessel collision risk. Reduced nighttime visibility shifts hotspots and elevates risk, daytime port activities create new high-risk zones, and severe weather drives vessels to typhoon shelters where higher density and poorer maneuverability increase danger. The proposed approach captures these shifts and yields an interpretable, actionable tool for collision risk assessment and maritime traffic management, supporting future maritime safety management.