Bond fatigue failure correlates with the increase in fatigue slip, influenced by the interfacial bonding properties of fibre-reinforced polymer (FRP) rebar and concrete under fatigue loading. Fatigue slip is a crucial indicator for estimating fatigue bond life. In this study, a comprehensive fatigue-slip dataset comprising 1,140 test results from published literatures was collected to develop predictive models using two tree-based learning algorithms: Random Forest (RF), and eXtreme Gradient Boosting (XGBoost). The dataset was categorized into 11 input parameters, including concrete properties, FRP rebar characteristics, and fatigue load conditions. To understand the influence of each parameter on fatigue slip and to identify the dominant bonding mechanisms, SHAP (SHapley Additive exPlanations) analysis was carried out. The analysis identified the top five contributing parameters, which were then used to derive a third-order polynomial fatigue-slip formula. Additionally, a theory-informed learning model was employed to predict fatigue slip by combining the shear-lag model and XGBoost model. The study further proposed a method for predicting the fatigue bond life based on these fatigue-slip prediction models, providing a unique insight into fatigue evaluation. The results demonstrated that the theory-informed learning model achieved better prediction accuracy for both fatigue slip and fatigue life.