Constructing three-way classifier with interval granulation neighborhood rough sets based on uncertainty invariance.

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Three-way decision with neighborhood rough sets (3WDNRS) effectively handles uncertainty in continuous data.
• Limitations of 3WDNRS include reliance on individual granules and the need for predefined thresholds.

Purpose of the Study:

• To address limitations of 3WDNRS by introducing interval granulation (IG).
• To develop an effective three-way classifier that enhances decision efficiency and model generalizability.

Main Methods:

• Proposed an interval granulation method based on DBSCAN.
• Developed an interval granulation neighborhood rough sets (IGNRS) model integrating IG with quality indicators.
• Constructed a three-way classifier (3WD-IGNRS) using the IGNRS model and minimum fuzzy loss.

Main Results:

• Conducted comparative experiments on 12 benchmark datasets against state-of-the-art granular-ball and classical machine learning classifiers.
• The proposed 3WD-IGNRS model demonstrated consistent performance improvements.
• Achieved an average accuracy improvement of 4.94% over the best granular-ball classifier.

Conclusions:

• The proposed interval granulation approach effectively overcomes the limitations of traditional 3WDNRS.
• 3WD-IGNRS offers a robust and more generalizable solution for decision-making with uncertain continuous data.
• The method shows significant potential for applications requiring efficient and accurate classification under uncertainty.