Unboxing machine learning models for concrete strength prediction using XAI

Area of Science:

• Civil Engineering
• Materials Science
• Data Science

Background:

• High-performance concrete is essential for durable infrastructure facing heavy loads and extreme weather.
• Accurate concrete strength prediction is key for optimizing performance, cost, and safety in construction.
• Machine learning (ML) offers advanced solutions for structural engineering challenges like concrete strength prediction.

Purpose of the Study:

• To evaluate and compare the performance of eight popular machine learning models for concrete strength prediction.
• To identify the most effective ML algorithm for predicting concrete strength based on mix design and loading conditions.
• To provide actionable insights for civil engineers using ML for concrete applications.

Main Methods:

• Evaluated eight ML models: Linear, Ridge, LASSO, Decision Trees, Random Forests, XGBoost, SVM, and ANN.
• Utilized a standard dataset of 1030 concrete samples for model training and testing.
• Employed SHAP (SHapley Additive exPlanations) for model interpretability.

Main Results:

• XGBoost, an ensemble learning technique, demonstrated superior performance.
• Achieved an R-Square (R²) of 0.91 and a Root Mean Squared Error (RMSE) of 4.37 with the XGBoost model.
• SHAP analysis provided insights into feature importance for the XGBoost model.

Conclusions:

• Ensemble learning methods, particularly XGBoost, are highly effective for concrete strength prediction.
• The study offers valuable data-driven insights for optimizing concrete mix design and construction practices.
• ML models, like XGBoost, can significantly enhance decision-making in civil engineering projects.