Evaluating the strength properties of high-performance concrete in the form of ensemble and hybrid models using deep learning techniques.

In the behavior of concrete, factors such as particle types, water content, aggregates, additives, and binders significantly influence its Compressive Strength (CS) properties. This study develops hybrid and ensemble models to predict compressive CS and slump flow of high-performance concrete (HPC) using a dataset of 191 mixtures. Admixtures like fly ash and silica fume enhance HPC through hydraulic or pozzolanic activity. Understanding the relationships between HPC components is crucial for computational analysis of CS properties. Deep learning techniques, including hybrid and ensemble methods, were developed to predict these properties with high accuracy. This paper focuses on forecasting models using T-SFIS, GBMBoost, and Decision Tree, combined with metaheuristic algorithms (GWO, QPSO) in hybrid and ensemble frameworks. Sensitivity analysis via SHAP and tenfold cross-validation evaluated model performance. Results showed that the GWO-based GBQP model achieved superior performance ([Formula: see text]=0.998, RMSE = 1.216 MPa for compressive CS). The ensemble DGT model ranked second, while T-SFIS performed lowest. For slump flow, TSQP excelled ([Formula: see text]=0.984, RMSE = 3.233 mm), closely followed by GBQP. These advanced techniques significantly enhance the efficiency and accuracy of predicting HPC CS properties.