Improving the quality of self-organizing maps by self-intersection avoidance.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Visualization

Background:

• The quality of self-organizing maps (SOMs) is crucial for effective data representation.
• Smooth SOMs provide clearer insights into complex datasets.
• Existing methods may produce maps with topological defects, hindering interpretation.

Purpose of the Study:

• To present a novel method for modifying the SOM learning algorithm.
• To enhance the quality of self-organizing maps by minimizing topology errors.
• To improve the clarity and interpretability of data visualizations generated by SOMs.

Main Methods:

• A modified learning algorithm for self-organizing maps is proposed.
• The approach focuses on preventing map self-intersections and near-intersections.
• This strategy aims to avoid states associated with low-quality map topology.

Main Results:

• The modified algorithm successfully reduces the number of topology errors in SOMs.
• Obtained maps exhibit improved quality and smoother data representation.
• Performance was validated using both synthetic and real-world data from diverse applications.

Conclusions:

• The developed method effectively enhances SOM quality by addressing topological issues.
• This approach offers a practical solution for practitioners seeking clearer data insights.
• The trade-off involves a potential increase in quantization error for improved topological integrity.