<p>Against the backdrop of global climate extremes and rapid urbanization, rail transit systems, as vital components of cities, are highly vulnerable to extreme rainfall. The novelty of this study lies in proposing a complete rail transit flooding risk assessment framework that integrates the Rail Transit Flooding Computational Model (RTFCM) and Machine Learning (ML) algorithms, including support vector machine regression (SVR), k-means clustering, and random forest (RF). Based on the simulation results of RTFCM for the flooding process of 31 entrances of Guangzhou Metro Line 18 under four historical extreme rainfall events, the SVR flooding model is trained. The trained SVR model is then applied to simulate the flooding process under various extreme rainfall scenarios. It is integrated with the K-means clustering algorithm to classify the flooding risk of the 31 entrances. Additionally, a sample dataset is created by incorporating 15 explanatory factors related to extreme weather, topography, anthropogenic, and rail transit, which are used to train the RF model for comprehensive prediction of flooding risk. The flooding risk distribution map reveals that high-risk areas in Guangzhou Metro Line 18 are primarily located in the Xiancun-Longtan and Nancunwanbo-Panyu square intervals. High-risk stations on the entire Guangzhou Metro line make up 27.9% of the total, with Line 2 containing the highest number of high-risk stations (13), and the APM line having the highest proportion of high-risk stations (77.8%). Among the 15 explanatory factors, distance to the road (DRO) and normalized difference built-up index (NDBI) are found to have the most significant influence on flooding risk. When compares to other ML models, the RF model demonstrates superior performance in both accuracy evaluation and validation against historical flood data. This more accurate and efficient approach can enhance the ability for flood prevention for metro systems and raise urban resilience.</p>