1) there isn't a clear and straightforward way to find the optimal value of k in k-means clustering 2) it is sensitive to the starting positions of the initial centroids