# Clusters of Catchbob! users

** Disclaimer: this post might be extra boring if you're not familiar with statistics and/or catchbob!**
Today I played a bit with principal components analysis and clustering techniques. The point was to categorize the behavior of CatchBob! users.

On this first picture we see the repartition of the users (each number is a user):

The principal components analysis gave me an interesting information, there seems to be 2 meaningful components to explain those data (component 1 and 2 explains a large proportion of the variance as attested by the following histogram). Time and path are the crux components (as shown on the previous picture).

Then I used two clustering techniques: *clara* and *pam* (partitioning around medoids). The results are pretty much the same.