Ensemble Fuzzy Feature Selection Based on Relevancy, Redundancy, and Dependency Criteria.

Area of Science:

• Computer Science
• Machine Learning
• Data Mining

Background:

• Classification systems face challenges with processing undesirable data, impacting performance.
• Filter-based feature selection methods aim to improve classification by selecting relevant features.
• Existing methods often assess feature significance individually or based on dependency, limiting comprehensive analysis.

Purpose of the Study:

• To introduce a novel ensemble feature selection method, fuzzy feature selection based on relevancy, redundancy, and dependency (FFS-RRD).
• To address the limitations of traditional methods by considering both individual and dependency feature discriminative abilities.
• To enhance the performance and stability of classification systems.

Main Methods:

• Developed the fuzzy feature selection based on relevancy, redundancy, and dependency (FFS-RRD) algorithm.
• Evaluated FFS-RRD by comparing it against eight state-of-the-art and conventional feature selection methods.
• Conducted experiments on 13 benchmark datasets using four well-known classifiers.

Main Results:

• The proposed FFS-RRD method demonstrated superior performance compared to existing feature selection techniques.
• Experimental results indicated significant improvements in classification performance across multiple datasets.
• FFS-RRD also showed enhanced stability in classification outcomes.

Conclusions:

• FFS-RRD effectively extracts comprehensive feature relations by considering relevancy, redundancy, and dependency.
• The proposed method offers a robust solution for improving the efficiency and accuracy of classification systems.
• FFS-RRD represents a significant advancement in filter-based feature selection for machine learning applications.