A clustering performance measure based on fuzzy set decomposition

Area of Science:

• Data Science
• Machine Learning
• Fuzzy Set Theory

Background:

• Clustering algorithms aim to reveal underlying data structures.
• Evaluating and comparing different clustering algorithm outputs is challenging.
• Existing methods lack a consistent way to rank partition quality.

Purpose of the Study:

• To develop a robust performance measure for assessing data partitions from various clustering algorithms.
• To establish a consistent ordering of data partitions based on their quality.
• To validate the proposed measure against classifier performance.

Main Methods:

• Utilized fuzzy set decomposition to define a performance metric.
• Applied the measure to evaluate partitions generated by diverse clustering algorithms on the same dataset.
• Compared the ranking from the performance measure with classifier error rates.

Main Results:

• The proposed performance measure successfully ordered different data partitions.
• The ranking provided by the measure showed consistency with classifier error rates.
• Demonstrated the effectiveness of the fuzzy set-based approach in evaluating clustering quality.

Conclusions:

• The developed performance measure offers a reliable method for comparing clustering algorithm outputs.
• This approach aids in selecting the most appropriate clustering for uncovering true data structures.
• The findings support the use of fuzzy set theory for enhancing clustering evaluation.