Multi-granularity collaborative clustering based on adaptive granular-balls.

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Granular-ball clustering enhances traditional methods but struggles with noise and boundary points.
• Existing methods require improvement for robust data representation and processing.

Purpose of the Study:

• To propose a novel multi-granularity collaborative clustering method using adaptive granular-balls (AGB-MCC).
• To address limitations of existing granular-ball clustering in handling noise and overlapping data points.

Main Methods:

• Adaptive granular-ball generation using a nearest neighbor approach.
• Classification of granular-balls into high-compactness and low-compactness subsets.
• Coarse-granularity clustering on high-compactness balls and fine-granularity refinement on low-compactness balls.

Main Results:

• AGB-MCC demonstrates superior performance in handling noise and overlapping points compared to state-of-the-art methods.
• The method shows strong adaptability across diverse synthetic and real-world datasets.
• High clustering robustness and computational efficiency were maintained.

Conclusions:

• AGB-MCC offers an effective solution for granular-ball clustering challenges.
• The proposed method advances the field by improving accuracy and efficiency in complex datasets.