Explainable artificial intelligence ( XAI )‐powered design framework for lightweight strain‐hardening ultra‐high‐performance composites ( SH ‐ UHPC )

Abstract Lightweight strain‐hardening ultra‐high‐performance concrete composite (SH‐UHPC) is an outstanding alternative for engineering applications and infrastructure thanks to its outstanding strength, toughness, ductility, and low density. The integration of artificial intelligence (AI)‐based modeling strategies into engineering problems can substantially accelerate material design processes while reducing experimental costs and time. Within this scope, the main motivation of this study is to predict the compressive strength (CS) of lightweight SH‐UHPC via a grey wolf optimization (GWO)‐integrated machine learning (ML)‐based modeling that offers high accuracy and reliability, thereby reducing experimental cost and time requirements while supporting environmental and economic sustainability. The overall results demonstrate that all developed GWO‐ML models achieved impressive performance levels in predicting the CS of lightweight SH‐UHPC. In particular, the GWO‐Extra Trees Regressor (GWO‐ETR) model demonstrated superior performance compared to other GWO‐ML models in terms of performance metrics (RMSE = 6.99, MAPE = 3.25%, and R 2 = 0.929), scatter plots (all points remained within a 10% margin of error), and uncertainty analysis ( U 95 = 10.0) during the testing phase. In addition, SHapley Additive exPlanations‐based feature importance analysis, as well as individual conditional expectation analysis and partial dependence plots, provided valuable insights and design suggestions for engineers and practitioners. Finally, an interactive graphical user interface has been developed to enable the application of similar data‐driven analyses on a larger scale and to obtain rapid predictions; however, it is suggested that the database be continuously updated to improve model performance and extend its generalization capacity in the future.