Evolutionary polynomial regression improved by regularization methods.

Area of Science:

• Geotechnical Engineering
• Data Mining
• Computational Intelligence

Background:

• Evolutionary Polynomial Regression (EPR) is a valuable data mining technique for geotechnical engineering challenges.
• Overfitting in EPR can compromise the effectiveness of testing datasets and reduce predictive accuracy.
• The fitness function is central to EPR's performance, necessitating methods to enhance its generalization capabilities.

Purpose of the Study:

• To investigate the application of L1 and L2 regularization methods to improve the EPR fitness function.
• To mitigate overfitting and enhance the generalization performance of EPR models in geotechnical contexts.
• To compare the predictive accuracy of EPR with classical, L1RM, and L2RM fitness functions.

Main Methods:

• Determined optimal regularization parameter (λ) values for L1RM and L2RM by analyzing test datasets.
• Implemented L1RM and L2RM within the EPR framework to modify the fitness function.
• Compared the regression development and prediction accuracy of standard EPR against L1RM- and L2RM-enhanced EPR.

Main Results:

• Regularization methods significantly improved the performance of the EPR fitness function.
• L1RM-enhanced EPR exhibited superior predictive accuracy compared to L2RM-enhanced EPR and standard EPR.
• The chosen regularization parameter values effectively addressed overfitting issues.

Conclusions:

• Integrating regularization methods, particularly L1RM, enhances EPR's predictive capabilities in geotechnical engineering.
• L1RM-enhanced EPR offers a robust solution for challenges related to construction constitutive models and engineering predictions.
• The study validates the effectiveness of regularization in improving data mining tools for complex engineering problems.