Hybrid optimization-driven AI framework for compressive strength prediction of hybrid fiber-reinforced recycled aggregate concrete

The optimization of fiber-reinforced high-strength concrete (FR-HSC) remains challenging due to complex nonlinear interactions among binder composition, fiber content, and mix proportions. This study proposes a hybrid optimization-driven AI framework integrating machine learning (ML), deep learning (DL), and hybrid optimization for accurate strength prediction and efficient mix design. A dataset of 392 samples, comprising 32 experimental and 360 literature-based data points, was developed covering a wide range of material compositions. Experimental results showed that the hybrid mix with 1.0% steel fiber and 0.45% polypropylene fiber achieved the highest 28-day compressive strength of 111.96 MPa, demonstrating the synergistic effect of fiber hybridization. Multiple ML models (SVR, RF, XGB-RR) and DL models (ANN, DCN, CNN-LSTM) were trained and optimized using a hybrid Genetic Algorithm–Bayesian Optimization (GA–BO) approach. Among all models, XGB-RR achieved the best performance with R² = 0.998 and RMSE = 1.397 MPa in training, and R² = 0.901 in testing, indicating strong predictive accuracy and generalization. SHAP analysis identified solution-to-binder ratio, steel fiber content, cement content, and superplasticizer dosage as the most influential parameters governing strength. An interactive GUI was also developed for real-time prediction and optimization. The proposed framework demonstrates high accuracy, interpretability, and practical applicability for sustainable FR-HSC mix design.