Machine Learning for Strength Prediction of Ready-Mix Concretes Containing Chemical and Mineral Admixtures and Cured at Different Temperatures

Ready-mix concrete (RMC) is one of the most utilized concrete products globally. To obtain a suitable prediction tool for performance of resulting RMC products can be beneficial to the whole industry. This paper presents a comparative study among five different machine learning algorithms [namely Multi-linear Regression (MLR), Support Vector Regression (SVR), K-Nearest Neighbors (KNN), Extreme Gradient Boosting (XGBOOST) and Deep Neural Network (DNN)] that can be used to predict the 28-day concrete strength of various mix designs and mixing conditions of 207 RMC mixtures. The RMC mix design parameters evaluated include cement content, coarse aggregate content, fine aggregate content, water content, addition of fly ash, slag, retarder, water reducer, air entraining agent, and the mixing process parameters include mixing time, number of revolutions and curing temperature. The machine learning algorithms were analyzed based on a 5-fold cross validation approach, and two particular algorithms were then designated for fitting the actual and predicted compressive strength data. Based on the comparison for 28-day compressive strength predictions, results reveal that the XGBOOST and DNN can be suitable strength predictors for RMC. The root mean square error (RMSE), Mean Absolute Error (MAE) and Coefficient of Determination (r(2)) score of the training data of XGBOOST approach were 1.73 MPa, 1.38 MPa, 0.925, respectively. The preciseness and distribution of the data analyzed by XGBOOST, offering it to be a suitable strength prediction tool for RMC.