Abstract This study presents a data‐driven analysis of the bond failure mode and bond strength between fiber‐reinforced polymer (FRP) bars and seawater sea sand concrete (SWSSC). A database compiled from experimental studies was employed to develop and evaluate six representative machine learning (ML) models, including K‐nearest neighbor, decision tree, random forest (RF), extremely randomized trees, extreme gradient boosting (XGBoost), and artificial neural network. Their predictive performance was systematically assessed using multiple evaluation metrics and further benchmarked against existing design‐oriented empirical models and code provisions. Furthermore, to enhance model interpretability and reveal the contribution of individual input features, the SHapley Additive exPlanations algorithm was employed to explain their decision‐making processes and offer insight into the best‐performing ML models. The results indicate that these ML models are capable of outputting higher precise predictions than traditional empirical methods. Particularly, the XGBoost and RF models exhibit excellent generalization ability and predictive accuracy in both classification and regression tasks. Furthermore, bond strength plateaus when rib depth exceeds 0.4 mm due to excessive radial tensile stress and reduced bar cross‐section. Similarly, bar strength beyond 800 MPa yields no further bond enhancement, as failure shifts from interface shear to concrete splitting. Bond length reduces bond strength via uneven stress distribution, especially beyond 100 mm. Concrete strength ranging from 35 to 75 MPa consistently enhances bond performance through improved load transfer and matrix capacity. Overall, this study not only demonstrates the superior predictive capability of ML‐based approaches but also provides physically interpretable insights into the bond mechanism, offering a reliable basis for the design and optimization of FRP‐reinforced SWSSC structures.
Wang et al. (Mon,) studied this question.