Computational Complexity and Its Influence on Predictive Capabilities of Machine Learning Models for Concrete Mix...

Area of Science:

• Concrete Technology
• Materials Science
• Artificial Intelligence

Background:

• Traditional concrete mix design methods struggle to meet modern demands for strength, eco-friendliness, and efficiency.
• Conventional approaches often lead to overengineering and difficulties in accurately predicting concrete properties.
• Machine learning (ML) offers a promising alternative for predicting concrete compressive strength in mix design.

Purpose of the Study:

• To investigate the relationship between the computational complexity of ML models and their accuracy in predicting concrete compressive strength.
• To evaluate the performance of deep neural network models with varying complexity levels for concrete mix design.

Main Methods:

• Five deep neural network models with different computational complexities were evaluated.
• Models were trained and tested using a large database of concrete mix designs and corresponding destructive test results.
• Performance was assessed using metrics like coefficient of determination (R²), mean squared error, and root mean squared error.

Main Results:

• A positive correlation was observed between increased model computational complexity and predictive accuracy.
• Higher complexity models demonstrated an increased R² and reduced error metrics (MSE, RMSE, etc.).
• This indicates that more complex ML models enhance the prediction of concrete compressive strength.

Conclusions:

• Increased computational complexity in ML models positively impacts their ability to predict concrete compressive strength accurately.
• These findings support the refinement of AI-driven methods for more efficient and precise concrete mix design.
• Future research can build upon these insights to further optimize ML applications in concrete technology.