Katlav, MetinTurk, Kazim2026-04-042026-04-0420262165-03732165-0381https://doi.org/10.1080/21650373.2026.2629850https://hdl.handle.net/11616/109215Ultra-high-performance concrete (UHPC) provides superior strength and durability but suffers from high cost and environmental impact. As a sustainable alternative, Ultra-High-Performance Geopolymer Concrete (UHPGC) requires reliable tools for predicting compressive strength (CS), yet existing frameworks remain limited, especially those combining AI, explainability, and experimental verification. This paper develops a Grey Wolf Optimizer (GWO)-enhanced machine learning framework to predict the CS of UHPGC using 179 mixes compiled from the literature. Four GWO-ML models (CatBoost, GBM, RF, ETR) were trained, with GWO-CatBoost achieving the highest performance (R2 = 0.971), followed by GWO-GBM (R2 = 0.967). SHAP-based analysis identified age, fiber, SF, SFL, and Na2SiO3 as the most influential variables. ICE and PDPs provided optimal design ranges for engineering use. A user-friendly GUI was also developed to predict CS along with cost and carbon footprint. Experimental tests on 10 new mixtures confirmed strong generalization of the GWO-CatBoost model (R2 = 0.884).eninfo:eu-repo/semantics/closedAccessUltra-high-performance geopolymer concrete (UHPGC)machine learningcompressive strengthexplainable artificial intelligence (XAI)Article10.1080/21650373.2026.26298502-s2.0-105030249622Q2WOS:001692018500001Q10000-0001-9093-7195