Abstract: In order to accurately predict the failure mode (rebar fracture and rebar pull-out) of grouted sleeves, a method driven by knowledge and data is proposed. Based on the test data of 348 cases of monotonically stretched grouted sleeves, the predicted value for grouting connection bearing capacity of grouted sleeve and the calculated value for ultimate tensile bearing capacity of rebar are obtained through machine learning regression algorithm and code formula respectively. The characteristic parameters of the relationship between them and failure mode are established based on domain knowledge, which is combined with the test data to form a new database. The failure mode is automatically predicted by using the new database and machine learning classification algorithm. The results show that the random forest algorithm has the best prediction effect on the bearing capacity and the failure mode. Compared with other prediction methods of the failure mode, the method driven by knowledge and data is more competitive in this paper. The general classification accuracy can reach 94.3%. The prediction results are interpreted by combining the methods of shapley additive explanations (SHAP) and partial dependent plot (PDP), and the influence of various parameters on failure modes is examined. Both the interpretive results of SHAP and PDP indicate that the anchorage length of rebar and the compressive strength of grouting material are the two test parameters that have the greatest influence on the failure mode. The interpretive results of PDP also give their critical values affecting the transformation of failure modes, which are 6.01d (d is the diameter of rebar) and 80.87 MPa, respectively.