Explainable data-driven formulation of chloride migration coefficient of eco-friendly concrete based on advanced automatic programming

Data-driven approaches have demonstrated high efficiency in designing concrete materials with a large design space for desired material properties, but conventional data-driven approaches are inconvenient for practitioners and lack interpretability. This paper presents an explainable data-driven approach to formulate chloride migration coefficients of concrete using Artificial Bee Colony Expression Programming and Gene Expression Programming, aiming to overcome the limitations of black-box machine learning models while achieving high accuracy. The proposed approach integrates concrete domain knowledge with data-driven techniques including feature selection, anomaly detection, and hyperparameter selection. The approach, implemented for both conventional concrete and green concrete that contains fly ash, slag, and silica fume, achieves Pearson correlation coefficients exceeding 0.97 for both conventional concrete and green concrete on training and testing sets, and exhibits Scatter Indexes of 0.1 for conventional concrete and 0.2 for green concrete, demonstrating high accuracy while providing explicit mathematical formulas for engineer-friendly applications.