0

I need some advise on curve fitting:

I have a 3 input function f(x,y,z). The function value should be non-negative.

I have some raw data from experiments (both input and output) based on which I'm trying to use Excel solver to find some coefficients of 2 order function of following format f(x,y,z)= (ax+by+cz)^2 +dx+ey+fz.

For the specific values of the experiment, the function is fine with the fitted data. However for data which is not as same as in experiments, I can receive negative values of the function.

Can you please suggest me what to do?

user1977050
  • 101
  • You have most likely done an overfitting of your data due to which you are getting a good model with the training data used to build you model but weird results with test data not used to build the model. Read about overfitting https://en.wikipedia.org/wiki/Overfitting – Nilotpal Sinha Oct 19 '16 at 07:11
  • I don't think that I've did overfitting as the sum of square errors is pretty high: 1.3, 4.5 (for 18 training points) – user1977050 Oct 19 '16 at 13:46

0 Answers0