Carbon fiber reinforced cementitious composite (CFRCC) possesses stable and controllable electrical properties for non-destructive structural health monitoring (SHM) by electrical resistivity measurement (ERM) to determine the durability of structures. It is necessary to build an accurate and robust prediction model of CFRCC's resistivity detected by ERM under the influence of complex factors to achieve better SHM performance. To address this problem, the present study developed an integrated modelling approach with multiple machine learning and optimization algorithms to identify the best model for predicting CFRCC's resistivity. A dataset containing 602 experimental instances was constructed based on information accessed in the research literature. Results show that the XGBoost model tuned by Bayesian optimization had the best predictive performance with the largest R2 scores (0.95 and 0.96) and was thus proposed as the prediction model. Weight scores of input factors were given by the proposed model revealing that CFRCC's resistivity highly depended on carbon fiber and sand content in the composite. The influences of critical factors on model output were further quantified by Shapley Additive Explanations (SHAP) algorithm. The developed method can be applied to aid the design optimization for CFRCC and the non-destructive SHM of CFRCC-based structures by ERM.