I have a number of metric and non metric variables. I want to use them as neural network input. Because of the relationship between some of them, I want to combine instead of eliminating them.
Thank you for giving me advice.
More info about the variables please. This largely relies on what vars you have and it's hard to provide a generic solution.
It's hard to say given the lack of detail. But you could run the numerical and non-numerical data as two separate models and concatenate their outputs and feed this into another model. Or you could combine the last hidden layer before the output of each and then feed that into your classifier. Model averaging is another method. There are different ways of creating an ensemble. You will have to experiment with different methods to see which fits best for your data.
- Badges
- Science topic