Silhouette Analysis for Performance Evaluation in Machine Learning with Applications to Clustering.

Area of Science:

• Machine Learning
• Data Mining
• Pattern Recognition

Background:

• Clustering is fundamental in machine learning for grouping similar objects.
• K-means and its extensions like kernel k-means are widely used for clustering.
• Kernel k-means offers non-linear clustering by projecting data into higher dimensions.

Purpose of the Study:

• To develop an unsupervised weighting function for clustering aggregation.
• To overcome the limitations of supervised methods like normalized mutual information (NMI).
• To create a more practical clustering approach by removing the need for training data.

Main Methods:

• Extending previous work on weighted majority voting for clustering.
• Utilizing the Silhouette index as an unsupervised criterion for weighting.
• Implementing a novel unsupervised weighting function based on the Silhouette index.

Main Results:

• The proposed method successfully aggregates clustering results without requiring a training set.
• The Silhouette index-based weighting function provides an unsupervised alternative to NMI.
• The new method is more aligned with the core concept of unsupervised clustering.

Conclusions:

• The developed unsupervised weighting function enhances clustering aggregation.
• This method offers a practical solution for scenarios where labeled training data is unavailable.
• The use of the Silhouette index makes the clustering approach more sensible and broadly applicable.