A survey of fuzzy clustering algorithms for pattern recognition. I

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Clustering algorithms model ambiguous, unlabeled data.
• Existing literature covers diverse approaches like fuzzy membership functions, learning methods, and network architectures.
• A unified framework is needed for comparing clustering systems.

Purpose of the Study:

• To propose a theoretical framework for comparing the expressive power of clustering systems.
• To establish a basis for comparison using common functional features.
• To unify the understanding of fuzzy clustering and soft competitive learning.

Main Methods:

• Reviewing existing clustering approaches: relative (probabilistic) and absolute (possibilistic) fuzzy membership functions, Bayes rule, batch/on-line learning, prototype editing, network architectures, and neuro-fuzziness.
• Proposing an equivalence between fuzzy clustering and soft competitive learning.
• Selecting functional attributes for comparing clustering algorithms.

Main Results:

• An equivalence is proposed between fuzzy clustering and soft competitive learning.
• A unifying framework for comparing clustering systems is established.
• A set of functional attributes for algorithm comparison is identified.

Conclusions:

• The proposed framework provides a basis for comparing diverse clustering systems.
• The equivalence highlights a common ground between fuzzy clustering and soft competitive learning.
• This work sets the stage for a systematic comparison of clustering algorithms in Part II.