Explainable Machine Learning for Multicomponent Concrete: Predictive Modeling and Feature Interaction Insights

Abstract

Multicomponent concrete is a widely used industrial material, yet its performance evaluation still relies heavily on expert judgment and long-term monitoring. With the rapid development of artificial intelligence (AI), machine learning has emerged as a promising tool in building science for analyzing complex datasets and reducing uncertainties associated with human factors. This study applies a variety of machine learning techniques—including linear and polynomial regressions, tree-based algorithms (Decision Tree, Random Forest, ExtraTrees, AdaBoost, CatBoost, and XGBoost), and the TabPFN model—to investigate the key factors influencing concrete compressive strength. To enhance interpretability, SHAP analysis was employed to uncover feature importance and interactions, offering new insights into the underlying mechanisms of multicomponent concrete. The findings provide a data-driven approach to support engineering design, facilitate decision-making in construction practice, and contribute to the development of more efficient and sustainable building materials.