Alternative Clustering using Feature Selection and Direct Clustering-Quality Maximization
Tao Thanh Vinh, 이종혁 · Open Access System for Information Sharing (Pohang University of Science and Technology) · 2012
We proposed a method for finding alternative clusterings of a dataset based on feature selection and direct maximization of clustering quality. We found the possible important features for the alternative clustering. We transformed the data using these features to make the original clustering not likely to be found while the alternative clustering is more likely to be found; we used a clustering algorithm that directly maximize the clustering quality so that the alternative clustering will be high quality. We tested our approach with some other approaches on our synthetic dataset, UCI datasets: Segmentation, Vehicle, Vowel, Ionosphere and Glass, and a textual dataset. Our approach was the most stable one as it resulted in the best JI and DI for most of the tests. Our results showed that feature selection and direct maximization of clustering quality are important for finding alternative clusterings.