Non-Exhaustive, Overlapping Clustering.

Area of Science:

• Data Science
• Machine Learning
• Network Analysis

Background:

• Traditional clustering algorithms like K-Means assume disjoint and exhaustive assignments, failing to address real-world data complexities.
• Existing methods often tackle data overlap or outlier detection separately, lacking a unified approach.

Purpose of the Study:

• To propose a unified objective function, NEO-K-Means (Non-Exhaustive, Overlapping K-Means), for handling both cluster overlap and non-exhaustiveness.
• To extend the NEO-K-Means concept to overlapping community detection in network analysis.
• To develop efficient optimization algorithms for the proposed objective function.

Main Methods:

• Introduced the NEO-K-Means objective function, a reformulation of K-Means with parameters for overlap and non-exhaustiveness.
• Extended the approach to weighted kernel K-Means for network community detection.
• Developed iterative and semidefinite programming algorithms for optimization.

Main Results:

• Demonstrated the effectiveness of NEO-K-Means in finding ground-truth clusterings with varying degrees of overlap and non-exhaustiveness.
• Showcased superior performance of the proposed method over state-of-the-art algorithms in overlapping community detection tasks.
• Validated the unified approach for addressing both overlap and outlier data points.

Conclusions:

• The NEO-K-Means objective function provides a unified framework for clustering complex, real-world datasets with overlapping and non-exhaustive properties.
• The developed algorithms efficiently optimize the objective, offering practical solutions for data analysis and network science.
• This approach advances clustering techniques by effectively managing data characteristics previously addressed in isolation.