Akademik Veri Yönetim Sistemi
Environment, Development and Sustainability, 2025 (SCI-Expanded, Scopus)
Fly ash–based geopolymer concrete (FA‑GPC) is emerging as a low‑carbon alternative to ordinary Portland cement (OPC) concrete, yet mix‑design choices that jointly maximize strength and minimize embodied CO2 remain serious. This study develops and tests a data‑driven framework that couples machine learning (ML) with metaheuristic optimization to address this trade‑off. A curated dataset of 220 FA‑GPC mixtures (16 inputs) was used to train five ML models, Decision Tree, Random Forest, back‑propagation neural network, XGBoost, and a Transformer, to predict 28‑day compressive strength (CS) and CO2 emissions; SHapley Additive exPlanations (SHAP) was applied for explainability. The Transformer achieved the best overall performance, with R2 = 0.83 and RMSE = 0.0637 when jointly predicting CS and CO2. SHAP ranked NaOH (dry) and superplasticizer as the most influential features. Three optimizers, Artificial Bee Colony (ABC), Particle Swarm Optimization (PSO), and Grey Wolf Optimizer (GWO), searched the feasible design space. GWO produced the best compromise: an optimal mix delivering 52.64 MPa 28‑day CS with CO₂ emissions of 0.018 kg/m3, whereas PSO and ABC yielded either lower CS (31.79 and 49.03 MPa, respectively) or higher CO2 (78.97 and 90.67 kg/m3, respectively). A practical GUI was built to allow practitioners to query the trained model and explore trade‑offs. These findings demonstrate that Transformer-guided GWO can effectively identify high-strength, ultra-low-carbon FA-GPC mixtures, paving the way for future advances in sustainable construction.