Hands-On Data Analysis with Pandas

Elbow point method for determining k

We picked eight clusters arbitrarily here, since this is the number of planets in our solar system. Ideally, we would have some subject area knowledge about the true groupings, or need to pick a specific number. For example, say we want to fit wedding guests at five tables so that they all get along, then our k is 5; if we can run three marketing campaigns on user groups, we have a k of 3. If we have no intuition as to the number of groups there will be in the data, a rule of thumb is to try the square root of our observations, but this can yield an unmanageable amount of clusters. Therefore, if it doesn't take too long to create many k-means models on our data, we can use the elbow point method. This ...