Hierarchical Topology-Based Cluster Representation for Scalable Evolutionary Multiobjective Clustering.

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Mining

Background:

• Evolutionary multiobjective clustering (MOC) algorithms offer advantages over single-objective methods, particularly when the number of clusters (k) is unknown.
• Large datasets pose significant computational challenges for MOC due to extensive search spaces and high fitness computation times.

Purpose of the Study:

• To propose a novel, scalable MOC algorithm that reduces computational overhead and enhances search efficiency.
• To introduce a hierarchical, topology-based cluster representation for improved MOC performance.
• To present a cluster ensemble strategy for determining final clustering results, applicable whether k is predefined or not.

Main Methods:

• A coarse-to-fine topological structure identifies seed points, and a tree-based graph represents clusters.
• A bipartite graph partitioning strategy with graph nodes facilitates effective cluster ensemble operations for generating offspring solutions.
• A cluster ensemble strategy is employed for final result determination, addressing the underexplored aspect in existing MOC methods.

Main Results:

• The proposed algorithm demonstrates superior clustering performance compared to existing methods.
• Significant improvements in computing efficiency were observed, especially for large-scale datasets.
• The hierarchical, topology-based representation effectively simplifies the search procedure and reduces computational burden.

Conclusions:

• The novel MOC algorithm offers a scalable and efficient solution for complex clustering tasks.
• The topology-based representation and cluster ensemble strategy enhance both the quality and speed of clustering.
• This approach provides a robust framework for MOC, particularly beneficial for large and high-dimensional data.