Modeling strength characteristics of basalt fiber reinforced concrete using multiple explainable machine learning...

Area of Science:

• Materials Science
• Civil Engineering
• Artificial Intelligence

Background:

• Machine learning (ML) models are increasingly used for predicting concrete strength.
• The 'black-box' nature of ML hinders result interpretation and user trust.
• Existing explainable AI (XAI) research in concrete often uses only one explanation method.

Purpose of the Study:

• To investigate the application and importance of XAI on ML models for BFRC strength prediction.
• To compare the explanations provided by two different XAI methods (SHAP and LIME) across multiple ML models.
• To enhance the interpretability and trustworthiness of ML predictions in concrete engineering.

Main Methods:

• Developed three tree-based ML models: Decision Tree, Gradient Boosting Tree, and Light Gradient Boosting Machine.
• Applied two XAI techniques, Shapley Additive Explanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME), to analyze model predictions.
• Predicted mechanical strength characteristics: compressive, flexural, and tensile strength of BFRC.

Main Results:

• Tree-based ML models achieved good accuracy in predicting BFRC strength characteristics.
• SHAP and LIME provided different feature importance rankings and magnitudes across the models.
• Explanations varied, indicating potential complexities in decision-making based on ML predictions.

Conclusions:

• XAI methods are crucial for understanding ML model behavior in concrete strength prediction.
• Discrepancies in XAI explanations highlight the need for further research and domain expert involvement.
• A user-friendly application was developed for rapid BFRC strength prediction.