Hi Julio!

Thanks for your kind words, I'm glad you liked it! I think they used hierarchical clustering (HC) for the panels, which is also a form of unsupervised machine learning. It is very similar to Kmeans, the difference is that you don't specify how many clusters you want to start with. HC does this automatically bottom-up starting with all the data points in there own cluster and every iteration merging data points that are similar, until all of them are merged together. Then you look at the Dendogram and choose an iteration with the amount of clusters you 'like'. I pandered a lot as to witch I should choose and ended up with Kmeans because it is very simple and easy to grasp and well known, and most important it runs very fast. HC doesn't run fast on bigger data sets unfortunately. In any case, all these unsupervised clustering methods are very similar and the results you get also.

https://en.wikipedia.org/wiki/Dendrogram

https://en.wikipedia.org/wiki/Hierarchical_clustering

Cheers!

Arie