Special Issue Article Open Access

Enhancement of Subspace Clustering For High Dimensional Data

Abstract

Clustering is a technique for grouping of similar objects among different objects from the dataset. Traditional clustering algorithms are more suitable for fewer dimensions. Feature selection can be used to select the relevant features from high dimensional dataset to improve the formation of clusters. In these methods the relationship between objects is not preserved since an object cannot be a member of more than one cluster. Subspace clustering is the solution to this problem, which preserves the relationship between objects. Semisupervised learning along with Subspace clustering uses the partial background knowledge which improves the cluster formation results. Constrained Laplacian Score is semi supervised based feature selection method for selecting the relevance feature. This method preserves the local and constraints ability among data objects. The combination of semi supervised clustering and feature selection enhances the subspace clustering process. Experimental results show that the semi supervised subspace clusters formed with semi supervised feature selection have good accuracy than the semi supervised subspace clusters formed only with the relevant dimensions.

To read the full article Download Full Article