A type-2 projected FCM

In the literature on subspace clustering, traditional clustering techniques have been extended for computing meaningful and interesting clusters in the appropriate subspaces of the high dimensional data. We present a novel algorithm to capture unobserved object relationships embedded in fuzzy subspaces. In order to model the uncertainties of fuzzy data, we propose a modification of fuzzy c-means algorithm through deployment of type-2 fuzzy sets for high dimensional data. By assigning membership grades to type-1 memberships, we are able to minimize the induced clustering error. The experimental results have shown that type-2 projected FCM efficiently prune the subspaces.