A Global-Relationship Dissimilarity Measure for the k-Modes Clustering Algorithm

Area of Science:

• Data Science
• Machine Learning
• Clustering Algorithms

Background:

• The k-modes algorithm is a standard for categorical data clustering.
• Existing dissimilarity measures in k-modes have limitations.
• Analyzing relationships between objects, cluster modes, and attribute differences is crucial.

Purpose of the Study:

• To propose a novel dissimilarity measure for the k-modes algorithm.
• To enhance the performance of categorical data clustering.
• To address limitations in existing k-modes dissimilarity measures.

Main Methods:

• Analysis of the k-modes algorithm and its dissimilarity measures.
• Development of a new dissimilarity measure named GRD.
• Experimental evaluation on four real-world UCI datasets.

Main Results:

• The proposed GRD measure considers both object-to-mode relationships and attribute differences.
• GRD demonstrated superior performance compared to two existing dissimilarity measures.
• Experimental results validated the effectiveness of GRD on real datasets.

Conclusions:

• GRD offers an improved approach to categorical data clustering using the k-modes algorithm.
• The novel dissimilarity measure enhances clustering accuracy.
• GRD provides a valuable alternative for categorical data analysis.