MGNR: A Multi-Granularity Neighbor Relationship and Its Application in KNN Classification and Clustering Methods

Area of Science:

• Machine Learning
• Data Mining
• Computational Intelligence

Background:

• Real-world data often has multiple granularities.
• Existing neighbor-based methods use a single granularity, limiting accuracy and efficiency.
• Manual setting of granularity in traditional methods is suboptimal.

Purpose of the Study:

• To develop a novel multi-granularity neighbor relationship approach.
• To improve the accuracy and time efficiency of neighbor-based machine learning algorithms.
• To address the limitations of single-granularity assumptions in data analysis.

Main Methods:

• Constructing a multi-granularity data representation using the granular-ball computing model.
• Leveraging this representation to create tailored, multi-granularity neighborhood relationships.
• Applying the enhanced relationships to K-Nearest Neighbors (KNN) classification and clustering.

Main Results:

• The proposed multi-granularity approach significantly boosts algorithm time efficiency.
• Tailored neighborhood relationships demonstrably improve algorithmic accuracy.
• Experimental results confirm the effectiveness of the multi-granularity neighbor relationship.

Conclusions:

• The novel multi-granularity neighbor relationship approach overcomes limitations of single-granularity methods.
• This method offers a more robust and efficient way to handle complex data distributions.
• The approach shows significant promise for enhancing KNN classification and clustering performance.