Predicting the Compressive Strength of Concrete by using Machine Learning Techniques

Abstract

Considerable efforts have been made to increase the strength of concrete by substituting some of the cement in the concrete with industrial waste, such as fly ash. Predicting the compressive strength of concrete, however, is a difficult undertaking since it depends on a number of elements, including the water-to-cement ratio and the size and form of the particles. The work on machine learning algorithms for Evaluating the strength of concrete with the inclusion of fly ash is presented in this publication. In order to determine the most accurate estimation technique, the study also compares the accuracy of various machine learning models, including ensemble models and non-ensemble models in predicting CS. For this purpose, a dataset with a diverse set of experimental findings, encompassing a broad spectrum of compressive strength values, was gathered from established literature and verified through statistical scrutiny. The amounts of cement, fine aggregate, coarse aggregates, fly ash, water content, percentage of superplasticizer, and curing days were the main input parameters for the ML models, while the output was the CS of concrete. The assessment of performance involved the utilization of various performance metrics, including MSE & R2 to assess accuracy and reliability. The comparison reveals that XGBoost Regressor, Bagging Regressor, and Random Forest are the most reliable models. Fly ash and other parameters’ effects on CS prediction were also fully understood thanks to the use of sensitivity and parametric analysis, which also helped to shed light on the relationship between input parameters and CS. The efficiency of machine learning methods is demonstrated in the study. It eliminates the need for expensive and time-consuming experimental experiments by providing researchers with a quicker and more economical way to assess how fly ash and other factors affect CS estimation.