Abstract: Differentiating damage states in corroded reinforced concrete (RC) components is crucial for the seismic resilience assessments of corroded RC structures. A machine learning-based multi-objective prediction method for drift ratio limits (DRLs) and failure mode (FM) in corroded RC beams is thusly proposed, supporting rapid damage state assessment of corroded components. A seismic test dataset of 110 corroded RC beams is established from existing literatures. The corrosion impact analysis on DRLs and fragility curve establishment for flexural and shear failure in corroded RC beams were conducted using the dataset. Sparrow search algorithm-extreme gradient boosting (SSA-XGBOOST) regression and classification models are developed and combined with recursive feature elimination with cross-validation (RFECV) to eliminate redundant features, and the optimal feature combination is determined. The Shapley additive explanations (SHAP) method are applied to enhance the interpretability of the SSA-XGBOOST "black box" model, revealing the complex mapping relationships between input features such as reinforcement corrosion degree indices, mechanical performance parameters, reinforcement parameters, and the output targets. A deep neural network (DNN) model incorporating classification and regression layers are constructed to achieve parallel multi-objective prediction. The DNN model aids in identifying seismic damage states in corroded RC components and in assessing the seismic resilience of corroded RC structures.