Ensembles based on random projections to improve the accuracy of clustering algorithms
Alberto Bertoni and Giorgio Valentini
Lecture Notes in Computer Science
We present an algorithmic scheme for unsupervised cluster ensembles,
based on randomized projections between metric spaces, by which
a substantial dimensionality reduction is obtained.
Multiple clusterings are performed on random subspaces, approximately preserving the distances between the projected data, and then they are combined using a pairwise similarity matrix; in this way the accuracy of each ``base" clustering is maintained, and the diversity between them is improved.
The proposed approach is effective for clustering problems characterized
by high dimensional data, as shown by our preliminary experimental results.