The growing hierarchical self-organizing map: exploratory analysis of high-dimensional data

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Mining

Background:

• Self-organizing maps (SOMs) are widely used for high-dimensional data analysis.
• Traditional SOMs have limitations due to static architecture and poor hierarchical representation.
• Addressing these limitations is crucial for advanced data mining applications.

Purpose of the Study:

• Introduce a novel Growing Hierarchical Self-Organizing Map (GHSOM) model.
• Develop a neural network that dynamically adapts its architecture during training.
• Improve the representation of hierarchical data relations for intuitive analysis.

Main Methods:

• Designed a hierarchical architecture composed of independent, growing SOMs.
• Implemented unsupervised training allowing architectural adaptation to input data.
• Incorporated global orientation for facilitated navigation across hierarchical branches.

Main Results:

• The GHSOM model demonstrates a problem-dependent, adaptive architecture.
• Successfully represents hierarchical data relations in an intuitive manner.
• Facilitates navigation and exploration of complex data structures.

Conclusions:

• The GHSOM overcomes limitations of static SOM architectures.
• Provides an intuitive method for exploring hierarchical data structures in data mining.
• Offers enhanced capabilities for explorative data mining applications.