I thought about: adding one linear layer summing an output from previous layer having 5 binary neurons and using mean square error.
I'm looking for similar problems, papers, tutorials, cost functions. Any suggestions?
[0 0 0 1 0]
I woluld try and encoding like this:
[1 1 1 1 0]
So the model would undestand that it is an increasing and somewhat correlated binary output.