Unsupervised Feature Selection for the $k$-means Clustering Problem