Interpretation of GAP statistic

Question

I´m having trouble deciding how to cluster my data based on this following analysis. I used clusGap in R, which gave me the following plot.

Provided I understand the whole thing correctly, I should be using 5 clusters based on this graph, because that´s the first k, where f(k) >= f(k+1)-SE(k+1). Right?

The problem is, the function itself sugests I use 1 cluster (ie no clustering), which I suppose makes sense, if you assume k=1 fulfills the first criterion too, given that f(k=2) is so very small. I´d happily use only one, too, but then NbClust suggests I use 2 clusters, no matter which precise method it uses.

As such, I´ve now completely lost my thread, and have no idea which clustering to use and why.

Answer 1

Plots like this do suggest to use k=1.

And that usually is correct. It means the data - or the way it was preprocessed and prepared - is unsuitable for clustering. When visualized e.g. with PCA, such data usually looks like one (hence k=1) big blob.