Machine learning-based models to predict waste-driven concrete’s compressive strength

Machine learning models for predicting concrete strength have shown limited accuracy in earlier investigations. This study aims to develop improved prediction models for compressive strength of sustainable concrete incorporating ceramic waste powder and stone dust. Using 513 mix design data from literature on concrete containing these waste materials, we evaluated multiple prediction models including standard and stepwise regression, Poisson regression, response surface regression along with artificial neural network models using Lavernberg-Marquardt and scale conjugate gradient algorithms. The neural network-based models showed superior performance in strength prediction, leading to detailed analysis of both algorithms. Sensitivity analysis using polynomial regression identified water-to-cementitious material ratio as the most influential parameter in strength prediction.