A novel causal inference method of exit choice behaviour analysis for passenger ships during emergency evacuation

Due to the complex internal layout of passenger ships, the large number of passengers, and the presence of designated assembly points, route selection during evacuation onboard ships has increasingly become a crucial factor affecting the overall evacuation process. However, variations in passengers’ risk perception and the heterogeneity of passenger groups often lead to marked differences in exit choice behaviour. To clarify the relationship between passengers’ exit choice behaviour and influencing factors, the Double Machine Learning (DML) method is employed in this study with optimisation of the nuisance function applied to identify key factors affecting exit choice from a causal inference perspective. First, 1,380 valid questionnaires are collected from passengers on ferry routes in the Bohai Bay area, covering essential dimensions such as individual attributes, behavioural preferences, and evacuation decision-making. Second, feature selection and model optimisation are conducted based on this dataset to construct a nuisance function with optimal average out-of-sample prediction performance. Finally, the DML approach is employed to conduct a causal effect analysis of exit choice behaviour, allowing for the systematic identification of key influencing factors. The findings indicate that alarm response and decision-making under congested conditions are identified by the DML as having significant causal impacts on exit choice. It is shown that relying solely on correlational analysis may lead to strategic misjudgements, whereas the application of causal inference enables more accurate identification of priority intervention targets.