An efficient and interactive feature selection approach based on copula entropy for high-dimensional genetic data.

Area of Science:

• Machine Learning
• Data Science
• Bioinformatics

Background:

• High-dimensional data presents challenges for machine learning models.
• Effective feature selection is crucial for improving model performance and interpretability.
• Existing feature selection methods may struggle with complex, high-order feature interactions.

Purpose of the Study:

• To propose an efficient and interactive feature selection approach using copula entropy.
• To develop a novel feature selection criterion based on the divisibility of multivariate mutual information.
• To enhance the stability and performance of the proposed method through an improved version (CEFS+).

Main Methods:

• Developed CEFS (Copula Entropy Feature Selection) using copula entropy to measure feature relevance and capture full-order interactions.
• Combined feature-feature and feature-label mutual information with a max-relevance min-redundancy strategy.
• Proposed CEFS+ incorporating a rank technique to address CEFS instability.
• Evaluated CEFS and CEFS+ using three classifiers on five diverse datasets.

Main Results:

• CEFS+ achieved the highest classification accuracy in 10 out of 15 evaluated scenarios.
• The proposed methods significantly outperformed six other commonly used feature selection techniques.
• CEFS+ demonstrated particular effectiveness on high-dimensional genetic datasets.

Conclusions:

• Copula entropy-based feature selection offers an effective strategy for high-dimensional data.
• CEFS+ provides a robust and accurate method for improving classification performance.
• The approach holds promise for applications in fields like bioinformatics where high-dimensional data is prevalent.