How to lower MSE in a function approximation problem?

I've got 2 matrices (X and Y). The matrix X: contains 10000 rows what means a 9 dimensional input. The matrix Y: This contains the output. So we'we got 10000 sample points.

(For example: for [10000, 9, 7842.2, 2588.8, 0.3301, 69.627, 1.0575e+06, 106.75, 4264.4, 74, 34.915] this input the output is 13.895)

This is my code so far:

%This is how I create a net
load('Givenmatrices.mat');
tx = x';
ty = y';
net = newff(tx,ty,100);
net = train(net,tx,ty);
save('net', 'net');

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%And I've got this funcion:
function y = funcApprox(x)
load net
tx = x';
y = sim(net,tx);
return

How can I get a resut with less MSE if I test it at random points? With this I get around 24.

[John D'Errico]

Please attach your code. Sending us to another site is not advised, and there is no reason to do so since attaching the code is so trivial.

Regardless, lowering the error may require that you choose a more intelligent model. For that, we cannot even know, since we do not see your data, nor know anything about the data, and what process generated it.

So if you wish a reasonable help, we would need to be given the data, as well as your code for modeling it, as well as any other information about the process that generated it.

[Vecsei Gábor]

Unfortunately I don't know anything about the given data. All I know is that the X matrix is 9x10000 and the Y is 1x10000. My teacher tests our solution in random points and than gives back the mse. So with this code I earned 17-16 mse in matlab but my teacher said that at random points it was 21 mse.