The bond between reinforcing steel and concrete governs anchorage and force transfer in reinforced concrete structures and deteriorates significantly when exposed to elevated temperatures. This study develops a prediction approach for the bond performance of reinforced concrete at elevated temperatures using experimental data from multiple studies. Machine-learning (ML) models, including Gradient Boosting (GB), eXtreme Gradient Boosting (XGBoost), K-Nearest Neighbour (KNN) and Decision Tree (DT) regressors, were developed and benchmarked against existing analytical equations and soft-computing approaches. The results show substantial improvements in predictive accuracy, with the GB model achieving R² = 0.99 and 0.97 for training and testing data, respectively. SHapley Additive exPlanations (SHAP) analysis was used to quantify the influence of input variables on the predicted bond strength (Tb). The results confirm the influence of temperature and geometric parameters on bond degradation and demonstrate the potential of explainable Artificial Intelligence (AI) to support structural fire assessment.