Interpretable Predictive Modelling of Basalt Fiber Reinforced Concrete Splitting Tensile Strength Using Ensemble...

Area of Science:

• Materials Science
• Civil Engineering
• Computational Mechanics

Background:

• Basalt fibers enhance concrete properties like strength and durability.
• High tensile strength basalt fibers significantly improve concrete's splitting tensile strength.
• Predicting concrete's splitting tensile strength is crucial for structural applications.

Purpose of the Study:

• To predict the splitting tensile strength of basalt fiber-reinforced concrete using ensemble machine learning.
• To evaluate the performance of Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), Random Forest, and Categorical Boosting (CatBoost).
• To analyze the influence of input features on prediction accuracy using SHAP and ICE plots.

Main Methods:

• Collected and curated experimental data on basalt fiber-reinforced concrete splitting tests.
• Applied ensemble learning algorithms: XGBoost, LightGBM, Random Forest, and CatBoost.
• Utilized performance metrics (RMSE, MAE, R²) and SHAP/ICE for analysis.

Main Results:

• XGBoost achieved a coefficient of determination (R²) greater than 0.9.
• All tested ensemble models demonstrated high predictive accuracy.
• SHAP and ICE plots provided insights into feature importance and model behavior.

Conclusions:

• Ensemble machine learning, particularly XGBoost, effectively predicts the splitting tensile strength of basalt fiber-reinforced concrete.
• The study validates the use of advanced computational methods for material performance prediction.
• Understanding feature impacts enhances the reliability and interpretability of predictive models.