I think that feature has relatively large sampling errors, making pretty much anything other than the total score not-so-reliable: I included the same picture twice by mistake and (for example) one copy has a large positive “girls 23-30” score and the other one has zero score in the same category. (On the other hand, I've just realised that the report keeps updating, as the differences between the two copies of the same picture are smaller than they used to be.)

Also, with a few exceptions the ranking I've got in the result matches the chronological order the pictures were taken, suggesting have become hotter and hotter over time. Or at least that my aesthetic preferences wrt photo portraits have improved.