An effective feature selection method via mutual information estimation.

Area of Science:

• Machine Learning
• Data Science
• Bioinformatics

Background:

• Feature selection is crucial for improving model performance and interpretability.
• Existing methods often struggle with datasets containing inter-feature dependencies.
• Mutual information is a powerful tool for quantifying feature relevance.

Purpose of the Study:

• To propose a new mutual information-based feature selection method.
• To address the challenge of feature dependency in selection processes.
• To evaluate the proposed method against established techniques.

Main Methods:

• A backward selection framework incorporating a novel mutual information criterion.
• Utilizing two established probability density function estimation methods for criterion computation.
• Comparison with existing mutual information and filter-based feature selection methods.

Main Results:

• The proposed method effectively identifies important features in datasets with inter-feature dependencies.
• Demonstrated superiority over benchmark methods in numerous artificial and real-world scenarios.
• Robust performance across diverse and complex datasets.

Conclusions:

• The developed feature selection technique offers a significant advancement.
• It provides a reliable solution for datasets with complex feature interdependencies.
• The method shows strong potential for application in various data analysis domains.