A Deep Generative Adversarial Network-Driven Framework with Hybrid Machine Learning Models for Predicting Split Tensile Strength of Fiber-Reinforced Recycled Aggregate Concrete

The prediction of split tensile strength (STS) of fiber-reinforced recycled aggregate concrete (FRRAC) is challenging due to limited experimental data and the complex interactions between fibers and recycled aggregates. This study presents a novel framework to address these challenges by integrating deep learning-based data augmentation with hybrid machine learning models. A deep generative adversarial network (DGAN) was employed to synthetically expand the dataset, effectively replicating the distribution and interrelationships of the original data, thereby ensuring reliability and overcoming the limitations of costly, time-consuming experiments. Two hybrid models, ANN-GWO and ANN-GTO, were then developed by coupling artificial neural networks (ANN) with advanced metaheuristic optimizers, gray wolf optimization (GWO) and gorilla troops optimization (GTO), respectively. Results showed that ANN-GTO achieved superior predictive performance with a coefficient of determination (R2) of 0.95 and root mean square error (RMSE) of 0.21. ANN-GWO also performed well, attaining an R2 of 0.89 and an RMSE of 0.40. To ensure transparency, SHapley Additive exPlanations (SHAP) and partial dependence plots (PDPs) were employed, highlighting key factors influencing STS. Finally, a user-friendly prediction interface was developed to enable instant estimation of STS values, making the framework practical and applicable for real-world construction scenarios.