A genetic graph-based approach for partitional clustering.

Area of Science:

• Data Science
• Machine Learning
• Computational Science

Background:

• Clustering is vital for data analysis, with growing interest in continuity-based methods over centroid-based ones.
• Spectral Clustering (SC) is a popular continuity clustering technique relying on graph cuts, but it is sensitive to parameter selection.
• Optimal parameter choice is critical for SC's clustering quality.

Purpose of the Study:

• To introduce a novel clustering algorithm that reduces parameter dependency compared to Spectral Clustering.
• To maintain or improve the quality of clustering solutions while enhancing robustness.
• To present an evolutionary approach for graph-based clustering.

Main Methods:

• Developed Genetic Graph-based Clustering (GGC), inspired by Spectral Clustering.
• Integrated a genetic algorithm (GA) to cluster the similarity graph.
• Employed evolutionary strategies to optimize the clustering process.

Main Results:

• The proposed GGC algorithm demonstrates reduced sensitivity to parameter choices.
• GGC shows increased robustness compared to standard Spectral Clustering.
• Experimental validation indicates competitive performance against classical clustering methods on synthetic and real datasets.

Conclusions:

• Genetic Graph-based Clustering (GGC) offers a robust and effective alternative for continuity clustering.
• The evolutionary approach successfully addresses parameter sensitivity issues in Spectral Clustering.
• GGC presents a promising method for data analysis requiring continuity-based clustering.