As I was getting nowhere in improving my model, I went back and re-read the competition intro.
It says that the image we have were scored by an expert, on a rank from 0 to 5. 
Are we supposed to have this feature somewhere? I can feel that it has potential to be golden… 
Hidden in the ID field perhaps? 
In any case, it would be interesting to compare the predictive power of our models, vs that single variable.
Hello @michael_bordeleau,
I think the feature is used to extract training labels, for example, it might be possible that the 0-5 scale feature is then scaled accordingly to 0-2 ( Normal, Pre-Alzheimers, post-Alzheimers) with some other data to get labels. It may be one of the reasons, i.e, fear of data leak in this competition. This is still my guess, I am not sure. Also thanks for your contributions, I learned a lot from your posts.
Jyot
