Finding the predictors of a neural network

Hello!

I have developed a neural network with its predictors (70 variables) and corresponding answers (5 variables). The thing is that now I want, from an answer, to find which predictors will generate that answer. I have already tried to make a neural network backwards changing predictors by answers but due to the complex data I have I get errors in this way. I have seen something about inverse propagation but I have no idea what would work in this case.

Please if anyone knows anything it would be very helpful.

Answer 1

[the cyclist]

Suppose each of your 70 predictors is a binary variable. That means that there are

2^70

ans = 1.1806e+21

possible combinations of predictors. For each of the 5 possible answers, you need to enumerate which of the 1e21 possible predictor combinations lead to it?

(It gets even worse if the predictors take on more values, or are continuous.)

Are you sure you need to do this?

Or maybe I misunderstand what you are asking for.

[FERNANDO CALVO RODRIGUEZ]

They are even more variable than that, so just imagine. Maybe half of them don't vary depending on the type, so maybe it comes down to 30 variables. I think I found the solution by performing the task with genetic algorithms with the objective function being the neural network.