Back propagation neural network

I trained some data of about 40 points having 2 inputs and 2 targets. After training, the results shows that there is a direct relationship between the output and the target. But the regression value is 1 for the test, training and validation set while the error difference is almost 0 or negative. How do i improve the network performance to get a more reasonable result?

Answer 1

[Greg Heath]

Insufficient info. Which net? How did you compute regression?

The typical measure of regression performance is mean-square-error.

However, MSE is better understood if it is normalized by the maximum possible value: the average target variance, MSE00

 E     = target - output           % error
 MSE   = mse(E)                    %
 MSE00 = mean(var(target',1))      % maximum possible mean-square-error
 NMSE  = MSE/MSE00                 & Normalized MSE:  0 <= NMSE <= 1

Hope this helps

Thank you for formally accepting my answer

Greg