If you are new with NNs stick with the defaults as much as possible. Input-Hidden-Output is sufficient for a universal approximator. It is very seldom that more is needed.
Although NEWFIT (regression) and NEWPR(classification) that call NEWFF (generic) are still available, they all are obsolete.
Do you have access to the current functions FITNET(regression) and PATTERNNET(classification) that call FEEDFORWARDNET (generic)? If so, choose one of the first two.
First practice on the MATLAB examples in the help and doc documentation.
First accept all defaults. If that doesn't work, increase the number of hidden nodes.
Don't hesitate to consult us again if you need help.
Hope this helps.
Greg
More examples can be obtained from
help nndatasets
doc nndatasets
Hope this helps.
Greg
Best Answer