This review presents a systematic review and SHAP-based interpretation of machine-learning (ML) predictions for reinforced concrete corbels (RCCs). A structured PRISMA-guided review was conducted using Scopus and Web of Science records, supplemented by reference checking. After duplicate removal and screening, 121 studies were retained for qualitative synthesis. The review discusses the effects of key parameters, including concrete compressive strength, span-to-depth ratio, corbel geometry, reinforcement, and loading history. Strengthening and repair studies were reviewed as a separate practical branch of the literature, while analytical and code-based models, including shear-friction, strut-and-tie, and triangular-truss approaches, were examined to clarify their assumptions and limitations. In the ML component, an XGBoost model was trained using 330 compatible vertical-load specimens with 13 min-max normalized input variables to predict the normalized experimental shear capacity, using an 80:20 training-testing split and a fixed random seed of 42. The model achieved R2 values of 0.9973 and 0.9090 and RMSE values of 0.0066 and 0.0388 for the training and testing sets, respectively; the gap between them indicates mild overfitting, so the model is used for interpretation rather than as a design equation. SHAP analysis was then applied to explain the influence of the main input variables. The results demonstrate that member geometry, a/d ratio, and concrete strength dominate the prediction trend, while reinforcement-related variables provide secondary but still meaningful contributions within the database range. The findings highlight the value of combining systematic review, code-model assessment, and explainable ML for understanding RCCs behavior.
