GeFeS: A generalized wrapper feature selection approach for optimizing classification performance.

Area of Science:

• Computer Science
• Machine Learning
• Bioinformatics

Background:

• Feature selection is crucial for improving classification accuracy and reducing model complexity.
• Existing methods may struggle with diverse dataset dimensions and sizes, and are prone to overfitting.
• Intelligent genetic algorithms (GAs) offer a powerful framework for optimization problems in machine learning.

Purpose of the Study:

• To propose GeFeS, a generalized wrapper-based feature selection method.
• To enhance the accuracy, robustness, and intelligence of genetic algorithms for feature selection.
• To effectively reduce feature dimensionality while improving classification performance.

Main Methods:

• Developed GeFeS, a wrapper-based feature selection approach utilizing a parallel intelligent genetic algorithm (GA).
• Introduced a new feature weighting operator and improved existing GA operators (mutation, crossover).
• Integrated nested cross-validation for robust model validation and employed k-nearest neighbor (kNN) for feature evaluation.

Main Results:

• GeFeS demonstrated effective generalization across datasets of varying dimensions and sizes.
• Achieved high average classification accuracies (e.g., 95.83% for lung cancer, 97.62% for dermatology).
• Significantly reduced feature numbers (e.g., from 56 to 28 for lung cancer, 279 to 135 for arrhythmia) while maintaining or improving accuracy.

Conclusions:

• GeFeS offers a robust and effective solution for feature selection in machine learning.
• The proposed intelligent GA enhancements contribute to superior performance and generalization.
• GeFeS successfully balances feature reduction with enhanced classification accuracy across diverse datasets.