I would like to ask which software do you use/recommend for creating recurrent types of networks (both, commercial and freeware)? I'm interested mainly in Elman and Jordan types.
here i will suggest you that better to work with decision support system working with ANN that is used in idrishi software especially for satellite images
My point is to build some black-box models of the river discharge time-series. I've found that in my case a multilayer perceptron gives much better results than linear multivariate regression. Target values are highly autocorrelated (first order autocorrelation is about 0,70-0,80 for other years).
So first, is my assumption that recurrental networks will cope with it better than feed-forward MLP is correct?
I don't have experience with any programming language, so I'd rather prefer some software based on GUI. I've tried trial version of the commercial NeuroSolutions software and I found it quite friendly for me. Did anyone of you used it?
For a typical time-series problem of river discharge like yours, i would recommend a recurrent neural network such as NARX or Neuro-Fuzzy(ANFIS). Both of these, based on my experience work really well with noisy data and if trained well, avoid overfitting of the nonlinear functions. You must control the performance of the network by checking the performance criteria. A lot of literature is available for this on the internet and i am attaching some for your aid. MATLAB would be a great platform for your case as it is easy to use and provides a simple GUI. The command for neural network usage is 'nnstart'