We are however using the same tools from the 90s.

Hey man, take this data, clean it up and load it in your Linux machine. Do some parameter optimisation and variable selection on a set of different models, among which also include a neural network. Then come back with a strategy to perform sensible model selection and comparison.

You can go back further if you remove the constraint on using Linux.

Tools were there, TBF I have seen MATLAB libraries from back then who were much better than your current scikits and the likes.