A novel two-stage feature selection method based on random forest and improved genetic algorithm for enhancing...

Area of Science:

• Machine Learning
• Data Science
• Computational Intelligence

Background:

• Advanced data acquisition leads to high-dimensional data, impacting machine learning model accuracy.
• Existing feature selection methods have limitations like incompleteness, instability, or inefficiency.
• Combining diverse feature selection techniques can overcome individual method shortcomings.

Purpose of the Study:

• To propose a robust two-stage feature selection method.
• To enhance machine learning classification accuracy and efficiency.
• To address limitations of single-method feature selection.

Main Methods:

• A two-stage approach utilizing Random Forest for initial feature ranking and elimination.
• An improved Genetic Algorithm with a multi-objective fitness function for global optimal feature subset search.
• Incorporation of adaptive mechanisms and evolution strategies to maintain population diversity and search efficiency.

Main Results:

• Significant improvements in classification performance across eight UCI datasets.
• Demonstrated excellent feature selection capability, reducing feature dimensionality effectively.
• Validated the efficacy of the combined Random Forest and improved Genetic Algorithm approach.

Conclusions:

• The proposed two-stage feature selection method effectively enhances machine learning classification performance.
• The integration of Random Forest and an improved Genetic Algorithm offers a superior alternative to single methods.
• This method provides a powerful tool for optimizing feature selection in high-dimensional datasets.