Predicting compressive strength of composite concrete materials by integrating machine learning techniques

Abstract

This study investigates the prediction of compressive strength in composite concrete materials made from industrial and agricultural by-products combined with conventional materials. The compressive strength of these composite materials is influenced by the ratios of the components used. Two machine learning techniques, Neural Network Regression Model and Ensemble Learning for Regression are employed to forecast the compressive strength based on experimental data from destructive tests on various composite mixes. The models are optimized using genetic algorithms and surrogate optimization methods for hyperparameter. The results of this study are significant in that they demonstrate the strong performance of both models in predicting compressive strength. The NNRM achieved a coefficient of determination (R²) of 0.9187, while the Ensemble Learning model outperformed with an R² of 0.9979. This high level of accuracy in the predictions highlights the potential of machine learning to effectively forecast material properties, allowing for better design and optimization of composite concrete materials. Such predictions can lead to significant practical benefits, such as more efficient use of raw materials and a reduction in the need for costly experimental testing. Moreover, improved predictions can enhance the durability and performance of composite materials, potentially reducing maintenance costs and increasing the lifespan of structures made with these materials. The study also emphasizes the importance of sensitivity analysis, which identifies key factors influencing compressive strength, enabling manufacturers to prioritize specific material properties for optimal results, this research not only contributes to the scientific understanding of composite material behavior but also has practical implications for cost savings, improved material durability, and the advancement of sustainable construction practices through optimized material design and performance prediction.