A Robust Three-Way Classifier With Shadowed Granular Balls Based on Justifiable Granularity

Area of Science:

• Machine Learning
• Data Mining
• Pattern Recognition

Background:

• Granular-ball (GB)-based classifiers offer adaptability but struggle with uncertain data instances.
• Current classification methods risk significant errors when faced with ambiguity.

Purpose of the Study:

• To develop a robust three-way classifier (3WC-SGBs) capable of managing uncertain data.
• To address the limitations of existing GB-based classifiers in handling ambiguity.

Main Methods:

• An enhanced GB generation method using information entropy and justifiable granularity.
• A shadowed mapping strategy to partition GBs into core, important, and unessential regions based on minimum uncertainty.
• Construction of a three-way classifier to distinguish certain and uncertain data instances.

Main Results:

• The proposed 3WC-SGB model demonstrates robustness in managing uncertain data.
• The classifier effectively mitigates classification risks associated with ambiguous instances.
• Comparative experiments show superior effectiveness and efficiency against benchmark and state-of-the-art methods.

Conclusions:

• The 3WC-SGB classifier provides a reliable approach for handling uncertain data.
• The method enhances classification safety by explicitly addressing ambiguity.
• The model offers a significant advancement in granular computing and classification.