Evidence accumulation clustering using combinations of features.

Area of Science:

• Machine Learning
• Data Mining
• Computational Statistics

Background:

• Ensemble clustering algorithms like Evidence Accumulation Clustering (EAC) are effective for arbitrary data shapes and cluster numbers.
• Clustering high-dimensional data with uninformative features presents a significant challenge for existing algorithms.

Purpose of the Study:

• To develop an enhanced Evidence Accumulation Clustering (EAC) algorithm for improved clustering of high-dimensional datasets.
• To address the challenge of uninformative features in large datasets by incorporating feature subsets into the clustering ensemble.

Main Methods:

• The enhanced EAC algorithm populates the clustering ensemble using clusterings derived from feature subsets.
• The method includes prewhitening of recombined data and weighting individual clusterings by their informativeness.
• An example implementation in Matlab is provided to demonstrate the algorithm's functionality.

Main Results:

• The enhanced EAC method demonstrates improved clustering effectiveness compared to the ordinary EAC algorithm on synthetic data.
• Weighting individual clusterings by a goodness-of-clustering measure enhances the overall evidence accumulation process.
• Clustering on feature subsets effectively mitigates the impact of uninformative features in high-dimensional data.

Conclusions:

• The proposed variant of EAC offers a robust solution for clustering high-dimensional data with potentially uninformative features.
• Feature subsetting, prewhitening, and weighted evidence accumulation significantly improve clustering performance.
• This enhanced EAC approach provides a valuable tool for data analysis in various scientific domains.