Improving projected fuzzy K-means clustering via robust learning