The novel hierarchical clustering approach using self-organizing map with optimum dimension selection

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• Data clustering is vital across various fields like healthcare and business.
• Current Self-Organizing Map (SOM) clustering often uses arbitrary dimensions, leading to suboptimal results.
• Existing methods frequently employ secondary algorithms for dimensionality reduction, which are not always optimal.

Purpose of the Study:

• To propose an optimized Self-Organizing Map (SOM) approach for data clustering.
• To determine the most effective higher dimension of SOM for a given dataset.
• To investigate the meaningfulness of the SOM weight matrix and discover optimal 2D configurations.

Main Methods:

• The proposed method utilizes SOM for both primary and secondary clustering stages.
• It focuses on selecting optimal higher dimensions for the SOM.
• The approach analyzes the SOM weight matrix to identify meaningful patterns and configurations.

Main Results:

• The optimized SOM approach achieved superior Adjusted Randomized Index scores on ten diverse benchmark datasets.
• Results outperformed existing web-based methods without attribute reduction.
• The method demonstrated effectiveness across medical, biological, chemical, and synthetic datasets.

Conclusions:

• Self-Organizing Map (SOM) is a robust clustering technique, performing comparably or better than methods like k-means.
• The proposed optimized SOM approach enhances clustering accuracy and efficiency.
• This method offers a superior alternative for clustering various data types.