• All algorithms require the user to specify the number of initial cluster centers.
• Some algorithms allow the user to specify the specific locations of the cluster centers or to choose the process by which the cluster centers are selected; (ENVI: no)
• Most simply assign the cluster centers (often with an undocumented procedure).
(ENVI: assigns pr evenly distributed in the data space)
• Clustering algorithms tend to be sensitive to the number and location of the initial cluster centers. Selecting a good set of seed values, can not only speed up the clustering process (faster convergence) but can also lead to "better" results.