Visualized analysis of mixed numeric and categorical data via extended self-organizing map

Area of Science:

• Data Mining and Machine Learning
• Artificial Intelligence
• Computational Statistics

Background:

• Real-world datasets frequently contain a mix of numeric and categorical attributes, posing significant analytical challenges.
• Existing self-organizing map (SOM) variants struggle with direct, unified handling of these mixed data types.

Purpose of the Study:

• To introduce MixSOM, an extended self-organizing map designed for the direct and unified analysis of mixed-type datasets.
• To improve the representation of cluster structures within mixed-type data on a self-organizing map.

Main Methods:

• Developed MixSOM, incorporating a data structure distance hierarchy for unified handling of numeric and categorical values.
• Implemented regularization of prototype distances proportional to map distances to enhance cluster structure portrayal.
• Conducted comparative experiments against Kohonen's SOM, generalized SOM, and visualization-induced SOM using synthetic and real-world datasets.

Main Results:

• MixSOM demonstrated superior performance in reflecting the inherent structure of mixed-type data compared to existing models.
• The model effectively facilitates further data analysis, including exploration at multiple levels of granularity.
• Experimental results validated the capability of MixSOM in handling complex mixed-type data scenarios.

Conclusions:

• MixSOM offers a robust and effective solution for analyzing mixed-type datasets, overcoming limitations of previous methods.
• The proposed model enhances data visualization and exploration by accurately representing underlying data structures.
• MixSOM provides a valuable tool for researchers and practitioners dealing with heterogeneous data.