How to overcome poor prediction of neural network.

I have created a neural network with nftool for prediction with default settings. Data used for input is matrix of 19x1140 i.e 19 input parameter and 1x1140 i.e 1 output parameter. The regression values comes around .99 and .98 but during prediction using sim(net,test) command it gives wrong prediction. How to solve this problem?

Answer 1

[Greg Heath]

Use datadivision and as few hidden nodes as possible. Repeat multiple times to get a good set of random initial weights.

What do you mean by prediction ... the test set performance?