Data Organization for Neural Network training

Hi, I'm trying to train a NARX network using time series data. I've got 80 sets of data I'd like to train the network with. For clarification, one set of data comprises of 6 financial indicators of X company as the input and the default probability of the company as the target, with 8 timesteps each. All in all I have the financial indicators and the default probability of 80 companies, hence 80 datasets. I would like to train the network to handle any time series of any company and then to do a one step prediction of the future default probability.

1. This is essentially a time series problem with independent samples. I'd really like some advice on how to organize my data for training as my limited understanding of Matlab's NN is that it can't train a network with independent sets of time series data.

2. For my mentioned purpose, does generating a script or the genfunction option in Matlab's GUI make a difference?

Thanks a lot!

Answer 1

[Greg Heath]

You don't have enough data to treat this as a timeseries example.

Better to treat it as a regression problem

 [ I N ] = size(input)  % [ 6 80 ]
 [ O N ] = size(target) % [ 1 80 ]

 help fitnet
 doc fitnet

Hope this helps.

Thank you for formally accepting my answer

Greg