Covering assisted intuitionistic fuzzy bi-selection technique for data reduction and its applications

Area of Science:

• Data Science
• Machine Learning
• Computational Intelligence

Background:

• Rapid data growth presents challenges like vagueness, uncertainty, redundancy, irrelevancy, and noise.
• Existing data reduction techniques struggle to address all these issues concurrently.
• Intuitionistic fuzzy (IF) and rough sets offer potential for handling uncertainty and vagueness.

Purpose of the Study:

• To develop a unified data reduction technique addressing vagueness, uncertainty, redundancy, irrelevancy, and noise simultaneously.
• To propose a novel method for simultaneous instance and feature selection in high-dimensional datasets.
• To enhance regression performance for specific applications like Antiviral Peptide IC50 prediction.

Main Methods:

• Development of a novel intuitionistic fuzzy (IF) similarity relation.
• Establishment of an IF rough set model based on the novel similarity relation.
• Presentation of an IF granular structure using the similarity relation and lower approximation.
• Utilization of IF granule importance for redundant size elimination and dimensionality reduction.
• Mathematical validation of proposed concepts and theorems.

Main Results:

• A comprehensive framework for simultaneous instance and feature selection is proposed.
• The method effectively eliminates redundancy and irrelevancy in both data dimensions and size.
• Vagueness is managed by rough sets, uncertainty by IF sets, and noise by IF granular structures.
• Experimental validation on benchmark datasets demonstrates the effectiveness of the proposed selection methods.

Conclusions:

• The proposed IF rough set-based methodology offers a robust solution for complex data reduction challenges.
• Simultaneous feature and instance selection significantly improves data quality for machine learning.
• The framework enhances regression performance, showing promise for drug discovery applications (e.g., Antiviral Peptide IC50).