Concrete Compressive Strength Prediction Using Boosting Algorithms

Abstract

This study proposes intelligent machine learning (ML)-based methods for concrete compressive strength prediction by utilizing a publicly available dataset. The methods employed are the XGBoost, CatBoost and TabNet algorithms. A total of 1030 data points are collected wherein the independent input variables are the amounts of the different components of the concrete mix design and the output variable is the compressive strength at different curing ages. The proposed boosting algorithm approaches are contrasted with a few other popular ML techniques used in this field, such as logistic regression, classification and regression tree, and artificial neural networks. It is found that XGBoost and CatBoost show significantly lower mean errors between predicted values and actual observations of the compressive strength than the contemporary architectures, while TabNet is not so efficient. TabNet’s lower efficiency of prediction can be attributed to the relatively small dataset that was used for this study.