Clustering Ensemble Model Based on Self-Organizing Map Network.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Single clustering algorithms often exhibit suboptimal performance, particularly with complex datasets.
• The Self-Organizing Map (SOM) is a powerful unsupervised learning technique for dimensionality reduction and visualization.
• Ensemble methods can improve the robustness and accuracy of machine learning models by combining multiple base learners.

Purpose of the Study:

• To propose a novel clustering ensemble method using a cascaded Self-Organizing Map (SOM) architecture.
• To address the limitations of single-clusterer performance by integrating multiple SOM networks.
• To enhance pattern recognition accuracy through a cascaded SOM approach.

Main Methods:

• A cascaded structure is introduced into the Self-Organizing Map (SOM) algorithm.
• Outputs from multiple SOM networks are combined as input to a subsequent SOM network.
• The method leverages high-dimensional data characteristics to mitigate individual SOM network errors.
• Random initialization of SOM parameters and training order eliminates the need for differentiated training samples.

Main Results:

• Experimental validation on classical datasets demonstrates the effectiveness of the cascaded SOM method.
• The proposed ensemble approach significantly improves pattern recognition accuracy.
• Accuracy improvements ranged from 4% to 10% compared to baseline methods.

Conclusions:

• The cascaded SOM clustering ensemble method offers a robust solution for improving pattern recognition.
• This approach effectively overcomes the performance limitations of individual SOM clusterers.
• The method provides a scalable and efficient way to enhance unsupervised learning tasks.