XGBoost algorithm-based prediction of concrete electrical resistivity for structural health monitoring

For structural health monitoring, electrical resistivity measurement (ERM) method is commonly employed for the detection of concrete's durability, as indicated by the chloride permeability and the corrosion of steel reinforcement. However, according to previous experimental studies, ERM results are susceptible to significant uncertainties due to multiple influencing factors such as concrete water/cement ratio and structure curing environment as well as their complex interrelationships. The present study therefore proposes an XGBoost algorithm-based prediction model which considers all potential influential factors simultaneously. A database containing 800 experimental instances composed of 16 input attributes is constructed according to existing reported studies and utilized for training and testing the XGBoost model. Statistical scores (RMSE, MAE and R2) and the GridsearchCV feature are applied to evaluate and optimize the established model respectively. Results show that the proposed XGBoost model achieves satisfactory predictive performance as demonstrated by high coefficients of regression fitting lines (0.991 and 0.943) and comparatively low RMSE values (4.6 and 11.3 kΩ·cm) for both training and testing sets respectively. The analyses of the attribute importance ranking also reveal that curing age and cement content have the greatest influence on ERM results.