Compressive strength of concrete material using machine learning techniques

Abstract

Significant efforts have been made to improve the strength of concrete by utilizing industrial waste like Fly Ash as a partial replacement of cement in the concrete. However, predicting the compressive strength of concrete is one of the challenging tasks since it is affected by several factors such as the shape and size of aggregates, water-cement ratio. The paper presents a study on the various investigation of machine learning (ML) algorithms to estimate the compressive strength (CS) of concrete containing fly ash (FA). The research also aims to compare the accuracy of different ML models, including non-ensemble models (Multiple Linear Regressor, Support Vector Regressor) and ensemble models (AdaBoost Regressor, Random Forest Regression, XGBoost Regressor, and Bagging Regressor), in predicting CS with a focus on identifying the most accurate estimation method. For this purpose, a dataset of 633 experimental results with a wide range of CS values, ranging from 6.27 MPa to 79.99 MPa, was collected from existing literature and validated using statistical analysis. The primary input parameters for the ML models included the quantities of cement, fine aggregate (FA), coarse aggregates (CA), fly ash (FA), water content, percentage of superplasticizer, and curing days, with the CS of concrete as the output. Performance evaluation was conducted using several performance indices, such as MAE, MSE, R2, MAPE, RMSE, and a20-index, to assess accuracy and reliability. The comparison reveals that XGBoost Regressor is the most reliable model, demonstrating the highest coefficient of determination (R2) of 0.95, a-20 index of 0.913, and the lowest RMSE value of 3.06 MPa and MAE of 2.13 MPa, while the Multiple LR model is the least accurate method with R2 equal to 0.52, a-20 index of 0.433, RMSE of 9.40 MPa and MAE of 7.68 MPa. Additionally, to provide deeper insights into the relationship between input parameters and CS, sensitivity and parametric analysis were employed, enabling a comprehensive understanding of the impact of fly ash and other parameters on CS prediction. From the study, it was observed that the age of the concrete is the essential feature, followed by cement and water, with information gain values of 32.91, 23.50, and 15.10, respectively. The study highlights the effectiveness of machine learning techniques, particularly the XGBoost Regressor, in accurately estimating the compressive strength of concrete. Furthermore, it offers researchers a faster and more cost-effective means of evaluating the effect of fly ash and other factors on CS estimation, avoiding the need for time-consuming and costly experimental studies.