A simple and fast method to determine the parameters for fuzzy c-means cluster analysis

Area of Science:

• Bioinformatics
• Computational Biology
• Data Science

Background:

• Fuzzy c-means clustering is vital for high-dimensional data analysis, including DNA microarrays and proteomics.
• A key limitation is the absence of rapid methods for optimizing algorithm parameters, risking inaccurate results.

Purpose of the Study:

• To develop a computationally efficient method for determining optimal fuzzy c-means parameters.
• To establish a direct functional relationship for setting the fuzzifier value based on dataset properties.

Main Methods:

• Applied fuzzy c-means clustering to randomized datasets to estimate optimal parameter values.
• Derived a functional relationship for the fuzzifier using dataset dimension and object count.
• Compared the minimum distance between centroids with other validation indices for cluster number estimation.

Main Results:

• The optimal fuzzifier value can be directly determined by a functional relationship, negating the need for extensive dataset evaluation.
• This approach improves cluster detection accuracy compared to using predefined fuzzifier values.
• The minimum distance between centroids is an effective and computationally efficient index for determining the optimal number of clusters.

Conclusions:

• A novel, fast method for optimizing fuzzy c-means parameters is presented.
• The proposed functional relationship for the fuzzifier enhances the reliability of clustering results.
• The minimum distance between centroids offers a superior alternative to computationally expensive validation indices.