Prediction of mechanical properties of recycled aggregate fly ash concrete employing machine learning algorithms

Abstract

This paper investigates the mechanical properties of concrete containing recycled aggregates and fly ash as a sustainable and eco-friendly solution for construction, which can reduce the demand for natural aggregates and cement, conserves natural resources, and curtails greenhouse gas emissions. Machine learning methodology can be a viable substitute for time-consuming and costly experimental procedures. This research aims to use machine learning techniques, specifically regression and classification tasks, to effectively forecast the mechanical characteristics of Fly Ash Recycled Aggregate Concrete (FARAC) with precision. The findings indicate that the algorithms are effective in providing precise predictions for the compressive and tensile strength as well as the slump of FARAC. Additionally, the influence of age on compressive strength tests has been investigated. The comparison of over five algorithms across all models, as mentioned above, indicates that Random Forest and XGBoost exhibited greater accuracy in regression and classification. The XGBoost algorithm demonstrated the highest level of accuracy for both compressive and tensile strength, achieving approximately 0.95. The random forest algorithms attained an accuracy of 0.86 for the slump. The Shapley Additive Explanation (SHAP) technique was employed to explain further the significance and value of the input variables that impact the mechanical properties.