Evaluation and estimation of compressive strength of concrete masonry prism using gradient boosting algorithm

Area of Science:

• Civil Engineering
• Materials Science
• Computational Mechanics

Background:

• Compressive strength (CS) is crucial for designing hollow concrete masonry structures.
• Traditional laboratory tests for CS are time-consuming and expensive.
• Machine learning offers a promising alternative for CS estimation.

Purpose of the Study:

• To develop and validate a Gradient Boosting (GB) model for predicting the CS of hollow masonry prisms.
• To compare the GB model's performance against other machine learning techniques.
• To identify key parameters influencing the CS of hollow masonry prisms.

Main Methods:

• Utilized a database of 102 hollow concrete specimens from published literature.
• Employed Gradient Boosting (GB) with K-Fold cross-validation and Particle Swarm Optimization (PSO) for hyperparameter tuning.
• Input features included mortar compressive strength (fm), block compressive strength (fb), height-to-thickness ratio (h/t), and fm/fb ratio.

Main Results:

• The GB model achieved high prediction accuracy (R²=0.977, RMSE=0.803 MPa, MAE=0.612 MPa, MAPE=0.036%).
• GB outperformed six other machine learning models in predicting CS.
• Sensitivity analysis revealed fb and h/t as the most influential factors on CS.

Conclusions:

• The developed GB model provides an accurate and efficient method for evaluating hollow masonry prism CS.
• This approach can significantly benefit practical applications in masonry structure design.
• The study highlights the potential of machine learning in materials engineering and structural analysis.