Abstract: To address the issue of evaluating concrete transmissivity, a machine learning prediction model for concrete electric flux was developed based on known mix proportion, revealing the key influencing factors and their effect law. Using an ensemble of six machine learning algorithms, including extreme gradient boosting (XGBoost) and support vector regression (SVR), prediction models were constructed from 48 experimental datasets. The shapley additive explanation (SHAP) function was employed to analyze feature contributions. The results indicate that the XGBoost model achieves the highest prediction accuracy (R²=0.983 6), with sand content and air content identified as the primary factors influencing concrete electric flux. Data analysis shows that the electric flux reaches its minimum value at a water-binder ratio of approximately 0.4, and a sand ratio between 30.5% and 35.3% can sustain electrical conductivity at a low level. This study provides a theoretical basis and quantitative method for predicting concrete durability.

Key words: concrete, electric flux, chloride ion, machine learning, feature importance analysis, big data