Assessment of Machine Learning Methods for Concrete Compressive Strength Prediction

This research sought to forecast concrete compressive strength through six machine learning (ML) algorithms namely Linear Regression (LR), Random Forest (RF), Decision Trees (DT), Gradient Boost (GB), Support Vector Machine (SVM), and Categorical Gradient Boost (CatBoost), and to examine the significance of the input factors on the concrete compressive strength. The study considered a wide range of literature data and examined the efficiency of boosted algorithms in predicting the strength of ordinary Portland cement concrete. A total of 1760 datapoints were gathered from the literature. In order to focus on the prediction of concrete without any pozzolanic content, the data points containing pozzolans were dropped, leaving 526 data points which were trained and tested on the selected ML algorithms. The model performances were evaluated based on mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE) and coefficient of determination (R2). An analysis was also performed to assess the sensitivity of the compressive strength to the input factors. Results showed that CatBoost, Gradient Boost (GB), Random Forest (RF) and Decision Trees (DT) had good performance, with CatBoost performing best with R2, MSE, RMSE, and MAE of 0.94, 17.663, 4.203, and 2.809 respectively, while the water-binder ratio showed the highest significance in affecting concrete strength.