Symmetry Based Automatic Evolution of Clusters: A New Approach to Data Clustering

Area of Science:

• Data Science
• Machine Learning
• Computational Intelligence

Background:

• Clustering algorithms aim to group similar data points.
• Existing methods often require a priori knowledge of the number of clusters.
• Developing robust clustering techniques that adapt to data characteristics is crucial.

Purpose of the Study:

• To introduce a novel multiobjective genetic clustering approach (MOLGC) utilizing line symmetry distance.
• To evaluate the performance of MOLGC against established clustering algorithms.
• To demonstrate the algorithm's ability to find optimal clustering solutions without pre-specifying the number of clusters.

Main Methods:

• The proposed algorithm, MOLGC, employs a multiobjective genetic approach.
• It uses both the Davies-Bouldin (DB) index and a line symmetry distance-based objective function.
• Nearest neighbor search is optimized using multiple randomized K-dimensional (Kd) trees.

Main Results:

• MOLGC successfully evolves near-optimal clustering solutions based on multiple criteria.
• The algorithm demonstrates superior performance across various cluster quality measures.
• Experimental results on artificial and real datasets confirm MOLGC's effectiveness compared to SBKM and MOCK algorithms.

Conclusions:

• The proposed MOLGC algorithm offers an effective approach for unsupervised clustering.
• Line symmetry distance provides a novel and efficient criterion for cluster assignment.
• MOLGC advances the field of genetic clustering by enabling adaptive and optimal solution discovery.