Assessing urban metro station flood resilience under extreme rainfall using an integrated machine learning–MCDM framework

Against the backdrop of climate change and rapid urbanisation, frequent extreme rainfall events have exposed urban underground rail transit systems to increasingly severe flood risks. Metro stations exhibit substantial differences in flood susceptibility, vulnerability, and recovery capacity, making it difficult for conventional risk assessment methods to accurately characterise their flood resilience. To address this challenge, this study proposes an integrated framework combining Positive–Unlabelled Learning (PU), the Best–Worst Method (BWM), and Criteria Importance Through Intercriteria Correlation (CRITIC) for assessing the flood resilience of metro stations (PU–BWM–CRITIC). The framework establishes a comprehensive indicator system comprising 23 indicators across three dimensions: flood susceptibility, vulnerability, and recovery capacity. Based on classical resilience-loss theory, an integrated resilience quantification model is developed to comprehensively evaluate resilience characteristics across the pre-disaster, during-disaster, and post-disaster stages. The proposed framework is applied to the Chongqing metro system as a case study. The results reveal pronounced spatial heterogeneity in the flood resilience of metro stations in Chongqing. Stations with low resilience are primarily concentrated in central urban areas and major transfer hubs, whereas highly resilient stations are more frequently distributed in peripheral urban areas. Although stations in central urban areas generally exhibit greater recovery capacity, their higher flood susceptibility and vulnerability contribute to greater potential resilience losses. Comparative experiments demonstrate that the PU–BWM–CRITIC framework maintains relatively stable station rankings and spatial patterns under different weighting schemes and extreme rainfall scenarios, indicating good robustness and adaptability across scenarios. The proposed framework provides a quantitative basis for identifying low-resilience stations, prioritising flood protection improvements, and allocating emergency resources.