need help for choosing topology of neural network

I want to create a neural network for pattern recognition. I have 10 different cases and i have made an output matrix that has 4 columns. I have 312 feature of different signals (I get them throw DFT and statistic formula).now i want to classify this signals to 10 groups. I simulate 500 observation for each type of signals (it means I have a 312*5000 matrix as an input and 4*5000 as output).which kind of neural network is better for this work?

Answer 1

[Greg Heath]

1. Use a 10*5000 target matrix with unit vector columns from eye(10)

2. Substantially reduce the input dimensionality (e.g., PLS, STEPWISEFIT, SEQUENTIALFS, PRINCOMP, ...).

3. Use PATTERNNET.

 help patternnet

 doc patternnet

Hope this helps

Thank you for formally accepting my answer

Greg

[mehdi]

thanks alot how can i choose the best number of neron and layer?

[Greg Heath]

Use one hidden layer. See my posts in the newgroup and answers. Use a subset of the following

greg Ndof Hub Hmax