Predicting the Mix Proportions of Concrete Containing Calcined Clay Using Hybridized CNN and XGB and Employing the Shapley Method for Sensitivity Analysis

The use of calcined clay as supplementary cementitious materials to blend with Portland cement has been increased since it resulted in many environmental advantages such as reducing CO2 emission. Moreover, compressive strength as one of the most important mechanical properties of concrete presents its performance. Concrete with good compressive strength needs a suitable mix design proportions, and computing its proportions is complex and time-consuming containing multiple phases. Recently, the application of machine and deep learning in prediction has shown promising results in civil engineering. Therefore, in this paper, predictive models have been developed by the use of a 1D convolution neural network (CNN) and extreme gradient boosting (XGB) to forecast the mix proportions of concrete. Also, to enhance the performance of CNN and XGB, the hyperband optimization algorithm as a novel hyperparameter optimization algorithm has been applied to be hybridized with the mentioned models. 714 sets of concrete are collected from published literature and used in this experiment. The error of both prediction models is low, and the average prediction accuracy ( \({R}^{2}\) $R^2$ score) is higher than 0.96. Furthermore, instead of using empirical methods of computing mixture design which can be complex and time-consuming, machine learning models can be developed and applied. Moreover, sensitivity analysis was applied to evaluate the effect of input features on output features.