Unsupervised Feature Selection for Multi-cluster Data via Smooth Distributed Score
Unsupervised feature selection is one of the key topics in data engineering. Previous studies usually use a score vector which has the same length as the feature number to measure the discriminating power of each feature, and the top ranked features are considered to represent the intrinsic multi-cluster structure of the original data. Among different algorithms, Multi-Cluster Feature Selection(MCFS) is one well designed algorithm for its superior performance in feature selection tasks. However, in practice the score vector of MCFS is often sparse, and it brings a problem that only few features are well evaluated about the discriminating power while most others’ are still ambiguous. In this paper, by simultaneously solving one L1-regularized regression and one L2-regularized regression, we propose a novel Multi-Cluster Feature Selection via Smooth Distributed Score(MCFS-SDS), which combines the two results to clearly evaluate the discriminating power of most features via smooth distributed score vector. It is extremely efficient when cluster number is small. Experimental results over various real-life data demonstrate the effectiveness of the proposed algorithm.
KeywordsUnsupervised feature selection clustering spectral regression
Unable to display preview. Download preview PDF.
- 1.Duda, R.O., Hart, P.E., Stork, D.G.: Pattern Classification, 2nd edn. Wiley-Interscience, Hoboken (2000)Google Scholar
- 2.He, X., Cai, D., Niyogi, P.: Laplacian Score for Feature Selection. In: Advances in Neural Information Processing Systems 18 (NIPS 2005) (2005)Google Scholar
- 5.Chung, F.R.K.: Spectral Graph Theory. Regional Conference Series in Mathematics, vol. 92. AMS (1997)Google Scholar
- 7.Cai, D., Zhang, C., He, X.: Unsupervised Feature Selection for Multi-Cluster Data. In: 16th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD 2010), pp. 333–342 (2010)Google Scholar