Unified Feature Selection Approach for Complex Data Based on Fuzzy β-Covering Reduction via Information Granulation

Area of Science:

• Data Mining and Machine Learning
• Artificial Intelligence
• Information Theory

Background:

• Feature selection is crucial for dimensionality reduction, model performance, and interpretability in data analysis.
• Existing methods often fail with complex real-world data, such as mixed feature types and missing values.
• A gap exists in feature selection techniques capable of handling heterogeneous and incomplete datasets.

Purpose of the Study:

• To propose a unified feature selection (UFS) approach for complex data environments.
• To develop novel uncertainty measures for fuzzy beta-covering.
• To address limitations of current feature selection methods in handling diverse data complexities.

Main Methods:

• Construction of monotonic uncertainty measures for fuzzy beta-covering from algebraic and information-theoretic perspectives.
• Design of two forward heuristic algorithms for fuzzy beta-covering reduction.
• Representation of complex, multi-feature data using fuzzy beta-covering via information granulation.

Main Results:

• The proposed UFS approach effectively handles complex data with multiple feature types and missing values.
• Experimental results demonstrate the superiority of the UFS approach compared to 12 state-of-the-art feature selection methods.
• The method shows significant improvements in performance and interpretability for machine learning models.

Conclusions:

• The developed UFS approach provides a robust solution for feature selection in complex data scenarios.
• The novel fuzzy beta-covering reduction techniques offer a powerful framework for data analysis.
• This work advances the field of feature selection by enabling effective processing of real-world, heterogeneous datasets.