A novel feature selection method and its application.

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Mining

Background:

• Feature selection is crucial for improving classification model performance and reducing dimensionality.
• Existing methods, particularly those based on rough sets, may not fully capture feature relevance.
• The need for methods that evaluate both definitive and uncertain feature relevance is recognized.

Purpose of the Study:

• To propose a novel feature selection method integrating rough sets and mutual information.
• To enhance the relevance and optimality of selected feature subsets.
• To validate the method's effectiveness across diverse classification tasks and a real-world medical dataset.

Main Methods:

• A novel feature selection approach combining rough set theory and mutual information.
• Feature selection guided by feature dependency and evaluated using a combined criterion of dependency and class-based distance.
• Utilizing mutual information to prune features that do not significantly enhance dependency.

Main Results:

• The proposed method achieves maximum feature subset dependency with a minimal number of features.
• The selected feature subsets demonstrate higher relevance compared to traditional rough set methods.
• Achieved 81.3% classification accuracy on a real Alzheimer's disease dataset, indicating near-optimal solutions.

Conclusions:

• The novel feature selection method effectively identifies highly relevant and near-optimal feature subsets.
• The integration of rough sets and mutual information offers a robust approach for dimensionality reduction in classification.
• The method shows significant promise for applications in medical data analysis and disease classification.