Yes, please, preferably starting from post 367 or 374 with some light clean up.
Taran, that means that you are not looking at all available data, may not have enough dimensions that you sort your points in, or, in case if you are using neural network, it is not deep enough to recognize all hidden dependencies.
There is no question that for both kids and adults there are several distinct groups with characteristic behaviors that will cluster together and also with some users falling in between.
However, if you select enough analysis criteria to sort the users in (dimensions) and your neural network is deep enough, then it should be able to recognize all clusters and cut through the noise.
I think there is enough raw data to make it much better. The trick is to find criteria that make more sense.