Clusters and Inertia?

[darfunkel]

Hi! Im new to this Forum.

I am just starting with hierarchical clustering (Agnes) and I would like to know more about inertia. What is the difference and/or relationship between variance and inertia? What is intra (group?) inertia and total inertia and how does it (or the changes of inertia as a function of the number of clusters) can help me choose the optimum number of clusters?

Somebody told me I can calculate the intra (group?) inertia as the squared distance between merging clusters, which can be obtained with R. Is this true?

Thank you very much in advance,
funk

[darfunkel]

Well, for those who might be interested, I found my answers.

Inertia is a measure of variation, or "spread". When you plot the explained inertia (intragroup sum of squares / total sum of squares) versus the number of clusters you usually notice an "elbow", a big curvature (or a peak in the second derivative) at some point. This is called the elbow criterion and it helps determine when it is no longer necessary to add more clusters. However, there may be many peaks and so this requires critical analysis.