Predicting the permeability and compressive strength of pervious concrete using a stacking ensemble machine learning...

Area of Science:

• Civil Engineering
• Materials Science
• Data Science

Background:

• Predicting pervious concrete performance is crucial but current models lack accuracy due to limited pore structure analysis.
• Existing models primarily use porosity, neglecting other vital pore characteristics, leading to prediction inaccuracies.

Purpose of the Study:

• To develop machine learning-based models for predicting the permeability and compressive strength of pervious concrete.
• To improve prediction accuracy by incorporating multiple pore structure parameters.

Main Methods:

• An ensemble model was constructed using six independent models, including multiple linear regression and the Stacking algorithm.
• Ninety pervious concrete specimens were prepared and tested to create an initial dataset, which was then augmented for model training.

Main Results:

• The integrated ensemble model achieved high prediction accuracy, with R² values of 0.925 for permeability and 0.928 for compressive strength.
• The ensemble model demonstrated a 15.9-23.9% improvement over single-model predictions, effectively overcoming limitations of traditional empirical formulations.
• Six input parameters proved effective for achieving high and straightforward model implementation.

Conclusions:

• The developed ensemble model significantly outperforms traditional empirical formulations in predicting pervious concrete permeability and compressive strength.
• Pervious concrete's compressive strength is more influenced by porosity variations than permeability changes.
• Machine learning offers a robust approach to enhance the predictive capabilities for pervious concrete properties.