Self-Adaptive Multiprototype-Based Competitive Learning Approach: A k-Means-Type Algorithm for Imbalanced Data...

Area of Science:

• Machine Learning
• Data Mining
• Unsupervised Learning

Background:

• Class imbalance is a well-studied problem, but imbalanced data clustering in unsupervised settings remains underexplored.
• Existing methods struggle with datasets where clusters have significantly different numbers of samples.

Purpose of the Study:

• To develop a novel method for imbalanced data clustering within the k-means competitive learning framework.
• To address the challenge of uneven sample distribution across clusters in unsupervised learning.

Main Methods:

• Introduced Self-Adaptive Multi-prototype-based Competitive Learning (SMCL) for imbalanced clusters.
• Employed multiple subclusters per cluster with automatic adjustment of subcluster numbers.
• Utilized a novel separation measure for subcluster merging and an internal validation measure to determine the final number of clusters.

Main Results:

• SMCL effectively handles imbalanced data clustering, inheriting advantages of competitive learning.
• The self-adaptive multiprototype mechanism accurately represents clusters of arbitrary shapes.
• The method automatically determines the optimal number of clusters for imbalanced datasets.

Conclusions:

• SMCL demonstrates efficacy in imbalanced cluster analysis compared to existing methods.
• The proposed approach offers a robust solution for unsupervised learning with imbalanced cluster sizes.
• Experimental results on synthetic and real datasets validate the effectiveness of SMCL.