Graph-Based Dissimilarity Measurement for Cluster Analysis of Any-Type-Attributed Data.

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Real-world data often contains heterogeneous attributes (numerical, nominal, ordinal).
• Defining similarity measures for heterogeneous data is challenging due to differing value characteristics.
• Clustering unlabeled data is valuable, but accuracy heavily relies on the chosen similarity metric.

Purpose of the Study:

• To propose a novel dissimilarity metric for clustering heterogeneous attribute data.
• To address the challenge of measuring similarity across diverse data types.
• To develop an effective clustering algorithm for mixed-attribute datasets.

Main Methods:

• Graph representations are built to model connections among heterogeneous attributes.
• A new dissimilarity metric is proposed, guided by these graph structures.
• A k-means-type clustering algorithm is developed incorporating the new metric.

Main Results:

• The proposed metric successfully computes dissimilarities between attribute values.
• The associated clustering algorithm handles arbitrary combinations of numerical, nominal, and ordinal attributes.
• Experimental results demonstrate the method's efficacy compared to existing approaches.

Conclusions:

• The novel dissimilarity metric and clustering algorithm are effective for heterogeneous data.
• This approach offers a robust solution for cluster analysis in diverse real-world applications.
• The graph-based metric provides a promising direction for future research in data clustering.