Machine learning and interactive GUI for concrete compressive strength prediction

Area of Science:

• Civil Engineering
• Materials Science
• Computational Science

Background:

• Concrete compressive strength (CS) is vital for structural integrity and design.
• Accurate CS prediction optimizes material usage, reduces costs, and minimizes trial-and-error.
• Machine learning (ML) offers advanced solutions for complex engineering predictions like CS.

Purpose of the Study:

• To enhance concrete compressive strength prediction using diverse ML models.
• To analyze the influence of eight key input parameters on CS.
• To develop a practical tool for efficient CS estimation.

Main Methods:

• Utilized 1030 experimental CS data points for training and testing ML models.
• Implemented both non-ensemble (regression, evolutionary, neural network, fuzzy-inference-system) and ensemble (adaptive boosting, random forest, gradient boosting) ML models.
• Conducted sensitivity analysis using Shapley-Additive-exPlanations (SHAP) and k-fold cross-validation for model assessment.

Main Results:

• The Categorical-Gradient-Boosting (CatBoost) model demonstrated superior predictive performance.
• CatBoost achieved a determination coefficient (R²) of 0.966 and a Root-Mean-Square-Error (RMSE) of 3.06 MPa.
• SHAP analysis identified concrete age as the most significant factor influencing CS prediction.

Conclusions:

• ML models, particularly CatBoost, provide highly accurate CS predictions.
• Concrete age is the primary driver of compressive strength.
• A Graphical User Interface (GUI) can facilitate rapid and cost-effective CS predictions for designers.