Cost-Sensitive Feature Selection by Optimizing F-Measures.

Area of Science:

• Machine Learning
• Data Science
• Computer Science

Background:

• High-dimensional data presents challenges for machine learning.
• Conventional feature selection methods often neglect class imbalance, leading to biased feature subsets.
• F-measure is a more suitable metric than accuracy for imbalanced datasets.

Purpose of the Study:

• To develop an effective feature selection algorithm that accounts for class imbalance.
• To optimize F-measures for imbalanced data by employing cost-sensitive learning.
• To ensure selected features adequately represent all data classes.

Main Methods:

• Decomposing F-measure optimization into a series of cost-sensitive classification problems.
• Generating and assigning class-specific costs based on theoretical guidance.
• Iteratively solving cost-sensitive feature selection problems to identify optimal features.

Main Results:

• The proposed cost-sensitive feature selection method effectively handles class imbalance.
• Selected features demonstrate improved representation of all data classes.
• Experimental validation on benchmark and real-world datasets confirms the method's effectiveness.

Conclusions:

• Cost-sensitive feature selection is crucial for imbalanced data.
• The proposed algorithm significantly enhances feature selection for imbalanced datasets.
• The method ensures selected features are representative and improve model performance.