Dimension selective self-organizing maps with time-varying structure for subspace and projected clustering

Area of Science:

• Data Mining
• Machine Learning
• Artificial Intelligence

Background:

• Traditional clustering algorithms struggle with noisy data and high-dimensional datasets.
• Subspace clustering is essential for identifying meaningful patterns in specific data dimensions.
• Noisy attributes can obscure true clusters, necessitating specialized approaches.

Purpose of the Study:

• To present an improved subspace and projected clustering method using self-organizing maps (SOMs).
• To enhance clustering quality, reduce computational cost, and simplify parameterization.
• To effectively identify the correct number of clusters and their relevant dimensions.

Main Methods:

• Development of a local adaptive receptive field dimension selective SOM.
• Introduction of a time-variant topology for improved adaptability.
• Application to both synthetic and real-world datasets for evaluation.

Main Results:

• The proposed SOM method demonstrates improved clustering quality and efficiency.
• It successfully identifies the correct number of clusters and relevant dimensions.
• Achieved nearly perfect results on synthetic datasets and outperformed previous methods on real-world data.

Conclusions:

• The enhanced SOM with time-variant topology offers a robust solution for subspace clustering.
• The method effectively handles noisy data and complex datasets.
• It represents a significant advancement in subspace clustering techniques.