Question
I used Neural Network fitting tool for training my data and got outputs for each target that i supplied to the network. Those outputs are well within the error range and give a good fit for the network. But, now i want to predict output based on input samples not included within the data set that i previously provided to the nnftool for getting the outputs. Please tell me how i can do that? The input samples are withing the training set range.
Related Questions
Expert Answer
Prashant Kumar
PhD Expert
Answered Aug 16, 2026
Incorrect understanding:
Generalization: Ability to perform well on nontraining data
Overfitting: Number of training equations, Ntrneq, not being sufficiently larger than the number of unknown weights, Nw, can be a cause of DECREASED generalization.
Mitigation: Either increase Ndof and/or use validation stopping(default) and/or use regularization (e.g., TRAINBR)
Insufficient information:
size(input) = [ I N ] = [ ? ? ]
size(target) = [ O N ] = [ ? ? ]
default number of training examples Ntrn = N-2*round(0.15*N) = ?
number of training equations Ntrneq = Ntrn*O
reference mean-square errors
MSEtrn00 = mean(var(trntarget',1)) % Biased
MSEtrn00a = mean(var(trntarget',0))% DOF adjusted
MSEval00 = mean(var(valtarget',1)) % Unbiased
MSEtst00 = mean(var(tsttarget',1)) % Unbiased
number of hidden nodes, H = ?
number of unknown weights Nw = (I+1)*H+(H+1)*O = ?
number of estimation degrees of freedom Ndof = Ntrneq-Nw = ?
normalized-mean-squuare-errors
SSEtrn = sse(trntarget-trnoutput)
MSEtrn = SSEtrn/Ntrneq % mse(trntarget-trnoutput)
MSEtrna = SSEtrn/Ndof
NMSEtrn = MSEtrn/MSEtrn00
NMSEtrna = MSEtrna/MSEtrn00a
NMSEval = MSEval/MSEval00
NMSEtst = MSEtst/MSEtst00
Have a different question? Ask here