Causal analysis of ship inspection data and maritime accidents through causal neural networks

Maritime transportation serves as the backbone of international trade, yet its growth is accompanied by an increasing number of maritime accidents. Therefore, reducing maritime accidents and mitigating their impact are crucial to supporting sustainable maritime industry development. Port state control (PSC) inspection is a key regulatory tool for enhancing maritime safety by inspecting foreign visiting ships to port. However, the actual impact of PSC inspection on reducing maritime accidents and how it can be further optimised remains unclear. This study employs a causal generative neural network model to explore the causal relationship among ship particulars, historical PSC inspection results, and future maritime accidents through a directed acyclic graph (DAG). By integrating causal discovery and causal inference, this study provides a robust framework for modelling maritime accident causation and overcomes limitations of traditional methods that often rely on discretisation or linear assumptions. Based on the identified optimal causal DAG, this study conducts decile-level causal intervention analyses on key contributing variables to quantify their effects on both the occurrence and severity of various types of maritime accidents. The results indicate that the optimal maximum PSC inspection interval is 409–509 days, which is consistent with the existing longest inspection time window (10–18 months) in the Tokyo Memorandum of Understanding. The probability and severity of maritime accidents exhibit a U-shaped relationship with the mean inspection interval, and the lowest risk occurs at 170–210 days. Additionally, the life-saving equipment conditions should be paid more attention in the PSC inspection. Due to the quantitative analysis nature, the new method will be able to help analyse the quantity of the increased or decreased risks against different types of maritime accidents, supporting a comprehensive perspective for safety management and accident prevention strategies.