Concrete Compressive Strength Prediction Using Machine Learning Algorithms

Abstract

The most widely utilized construction material is concrete. Concrete's physical qualities differ depending on the kind. In this paper, we predicted the compressive strength of four types of lightweight aggregate geopolymer concretes (LWAGC), namely, lightweigh expanded clay Leca, recycled foam masonry aggregate RFA, Porcelanite aggregate PA and recycled brick aggregate RBA. For predictions, we used seven models, specifically, Artificial Neural Network (ANN), Convolutional Neural Network (CNN), Long Short Term Memory (LSTM), Decision Tree (DT), Random Forest (RF), Support Vector Regressor (SVR) and Linear Regression (LR). For evaluation, we employed six metrics, Root Mean Square Error (RMSE), Mean Squared Logarithmic Error (MSLE), Mean Square Error (MSE), Mean Absolute Error (MAE), Root Mean Squared Logarithmic Error (RMSLE) and Coefficient of Determination R². The results of predictions demonstrate that RF offers high accuracy for Leca type (MSE= 2,6697; RMSE= 1,6339; MSLE= 0,0072; RMSLE= 0,0851; MAE= 1,3434; R²=0,8687), Porcelanite type (MSE= 2,9650; RMSE= 1,7219; MSLE= 0,0065; RMSLE= 0,0805; MAE= 1,5693; R²=0,8437) and RFA type (MSE= 1,7028; RMSE= 1,3049; MSLE= 0,0059; RMSLE= 0,0768; MAE= 1,1300; R²=0,8099) because the predictions are closer to the real values, and DT offers better predictions than other models for RBA type (MSE= 3,3069; RMSE= 1,8185; MSLE= 0,0066; RMSLE= 0,0812; MAE= 1,5610; R²=0,8764). İn order to test the models in predicting new value we gave them a 675° as new temperature and we found that that LR is more accurate than other models in predicting the CS of Leca, RBA and Porcelanite types. However, CNN outperformed other models in predicting RFA type.