Conductive mortar auxiliary anode (CMAA) is one of the main materials to realize stable operation of cathodic protection system. However, the conductive material in CMAA is complex and diverse, and its basic physical properties have a complex functional relation with the type and content of conductive material. At present, there is no uniform evaluation criterion, which can realize the evaluation of the action properties of different conductive substances. To address this issue, machine learning was proposed to establish predictive models between input and output factors. The SHAP-value was used to analyze the model's global and local features, and the partial dependency of prediction results. The results show that: on the basis of 241*3 groups of databases, the predictive performance analysis of RFR, GB, XGBoost and LGBM models shows that LGBM is the optimal model, and the determination coefficients of its training set and test set are 0.98 and 0.97, respectively. The LGBM model, combined with SHAP-Value analysis and PDP analysis, reveals the importance ranking of the input features that affect the compressive strength, and can also provide a quantitative evaluation of the effect of the input factors against the compressive strength, providing a new analysis method for material design and performance optimization.