The corrosion-induced failure of marine reinforced concrete structures, stemming from prolonged exposure to environments with elevated chloride content, poses a significant threat to structural integrity and serviceability. Traditional models predominantly depend on prior information to estimate the spatial stochastic degradation of aging structures, leading to a lack of comprehensiveness and accuracy in evaluation results. To address these limitations, a data-driven probabilistic assessment model for corrosion failure is proposed. This model incorporates a pre-training phase that integrates electrochemical simulations with long-term exposure data to establish the spatiotemporal distribution of corrosion failures, thereby facilitating a thorough assessment of overall damage by quantifying the corrosion failure proportion while propagating epistemic uncertainty. Subsequently, a multivariate function is derived using the non-dominated sorting genetic algorithm to elucidate the relationship between input and output variables. The efficacy of the data-driven model is evaluated through experimental observations and model predictions. The Hangzhou Bay Bridge project serves as a case study, wherein this model is demonstrated using actual data obtained from exposure experiments and field detection. Furthermore, the proposed model enables a rapid prejudgment of corrosion failure proportion, significantly reducing diagnostic time compared to conventional procedures.