Machine learning models for mechanical properties prediction of basalt fiber-reinforced concrete incorporating...

One of the most significant advancements in basalt fiber (BF) technology is its application in Basalt fiber reinforced polymers (BFRP). The production of BFRP utilizes basalt rock, a naturally abundant resource, resulting in a composite that generates approximately 74% less carbon emissions compared to traditional steel, aligning with global sustainability goals. This study employs previously published experimental datasets on basalt fiber-reinforced concrete (BFRC) to statistically predict compressive strength (CS) and splitting tensile strength (STS) using advanced machine learning. The dataset includes 270 CS and 267 STS samples, split into 70% training and 30% testing, enabling accurate, data-driven predictions without the need for new laboratory experiments. In addition to the parametric analysis of Shapley additive explanation (SHAP), machine learning models were used, namely support vector regression (SVR), random forest regression (RFR), decision tree (DT), bagging regressor (BR), and gradient boosting regression (GBR) with grid search hyper-tuning. Additionally, the model generated SHAP interaction plots to show the impact of each characteristic on an individual prediction. The results found that GBR model performance is the most precise prediction of compressive strength compared to other models, achieving an R² of 0.99 for training phase and R² of 0.86 for testing phase. But SVR model outperforms the other four models in STS prediction, with the coefficient of determination (R²) value of 0.99 during the training stage and R² of 0.97 for the testing stage. The Shapley additive explanations (SHAP) method was used to display the effect of each input parameters on model prediction. The cement and silica fume were found to have the highest positive influence on BFRC compressive and tensile strength. The basalt fiber (BF) diameter as an input parameter was found to have the highest effect on STC. Finally, the concrete designers can now easily and affordably predict CS and STS using a graphical user interface, without conducting expensive computations or experiments.