An interpretable surrogate modeling framework for rice husk ash concrete using copula-based virtual sampling

This study proposes an interpretable and variance-aware sensitivity analysis framework for analyzing the compressive strength of rice husk ash concrete under correlated input conditions, using statistically consistent virtual sampling and surrogate-based sensitivity analysis. A Gaussian copula is used to construct a virtual design space that preserves empirical marginal distributions and inter-variable dependencies. An Extreme Gradient Boosting surrogate optimized using the Covariance Matrix Adaptation Evolution Strategy achieves high predictive accuracy on unseen data (mean R² = 0.974). Global variance attribution is performed using a Sobol-inspired, permutation-based sensitivity measure designed for correlated inputs, enabling robust sensitivity analysis of black-box models without relying on the independence assumption. Cement, Water, and Age are identified as dominant contributors, with notable interaction effects between Cement-Water and Coarse aggregate-Superplasticizer. These findings are further supported by SHAP analysis and partial dependence plots. External validation using independently published experimental data and deployment in a web-accessible environment demonstrate the framework's reliability and practical relevance for sustainable concrete mix design. Overall, the proposed framework offers a practical and interpretable tool for supporting robust and sustainability-oriented concrete mix design under correlated input conditions.