suppose we wish to run k-means clustering on our data, but we have too many features upfront. what kind of pre-processing can be done to improve the performance of k-means?