Hands-On Data Analysis with Pandas

Evaluating clustering results

The most important criterion for evaluating our clustering results is that they are useful for what we set out to do; we used the elbow point to pick an appropriate value for k, but that wasn't as useful to us as the eight clusters. That being said, when looking to quantify the performance, we need to pick metrics that match the type of learning we performed.

With the planets data, we performed unsupervised clustering because we don't have labels for what each cluster should be, and therefore we can't measure how well we did against those. This means that we have to use metrics that evaluate aspects of the clusters themselves, such as how far apart they are and how close the points in a cluster are together. ...