Developing machine learning models to predict the fly ash concrete compressive strength

Abstract

The advent and progress of machine learning (ML) have profoundly influenced civil engineering, especially in forecasting concrete's mechanical properties. This research focuses on predicting the fly ash (FA) concrete compressive strength (CS) using six different ML models: linear regression (LR), decision tree (DT), random forest (RF), extreme Ggradient boosting (XGB), support vector regression (SVR), and artificial neural network (ANN). A dataset comprising 1089 records, each with 12 input features, including the chemical compositions of FA, was used to train these models. The models' performance was assessed and compared using mean square error (MSE), mean absolute error (MAE), and the coefficient of determination (R2), with validation achieved through the K-fold cross-validation method. Among all the models evaluated, XGB was the most accurate, attaining an R2 value of 0.95. To interpret and understand the ML model predictions, Shapley Additive Explanations (SHAP) analysis was employed. It revealed that curing days, water-binder ratio, cement content, and superplasticizer are the most critical factors in predicting the FA concrete CS. These results indicate the potential of ML models, especially extreme gradient boosting, in accurately predicting concrete strength, promoting more efficient and effective use of FA in construction. Additionally, a graphical user interface (GUI) was created to enhance user interaction with the prediction models, improving the utility and accessibility of ML applications.