Machine learning-driven predictions of electrical resistivity for smart structural health monitoring in conductive cementitious composites

The optimization of electrically conductive cement-based composites (ECBCs) is crucial for developing intelligent infrastructure with self-sensing capabilities. This study presents a rigorous comparative analysis of nine machine learning algorithms for the high-accuracy prediction of electrical resistivity in ECBCs. We evaluated models including XGBoost, Support Vector Regression, Linear Regression, and K-Nearest Neighbors based on predictive accuracy, computational efficiency, and robustness to noise. Among them, the XGBoost algorithm demonstrated superior predictive performance, achieving the highest R² value of 0.9803 alongside the lowest MAE and RMSE values of 333.72 and 1116.41, respectively. Tree-based ensemble algorithms collectively achieved superior predictive performance across all evaluation metrics, while linear models exhibited comparatively poor performance. Although the KNN algorithm also showed high accuracy (R² = 0.9781), it incurred the highest computational latency due to its instance-based learning paradigm. Notably, KNN displayed exceptional robustness against Gaussian noise, maintaining an R² of 0.9565 even at a 20% noise level, attributed to that the introduced noise did not severely disrupt the neighborhood structures. The influence of key input variables was quantitatively interpreted using SHapley Additive exPlanations (SHAP) and gain-based importance analysis. This work demonstrates significant potential for practical application of ECBCs and paves the way for data-driven design of multifunctional construction materials.