To base a model on party loyalty when early voting records for 18-29 year olds are being broken all over the States suggests that you're starting with a flawed premise.

Roughly a third of the 18-29 cohort will be first time voters and most of the 2016 records are being smashed by factors much greater than three.

In addition, 2016 had extremely low turn out, whereas almost half the number of the votes cast in 2016 have already been cast.

I'd say the fact that both the youth vote and high turnout lean Democrat will be factors too great for your current model to manage.