Non-linear multivariate regression. Short answer needed.

Hi

I want to do a multivariate regression, finding for values A, B, C in the following formula Y=(A*x*z^B)/( x*z^B+C)

given that I know Y-values connected to x and Z values. Is there a function in MATLAB that makes it possible with a reasonable amount of work? Thanks

/ Jocke

[John D'Errico]

Short answer given - Yes.

Answer 1

[John D'Errico]

The long answer is, it depends on whether this is the correct model for your data. It depends on whether you can find reasonable starting values for your coefficients. It depends on the quality of the data. Is there a great deal of noise here? Exponents (B) are sometimes problematic to get good convergence in a model, and noise will cause difficulties. Do you have sufficient data. (Nobody ever has as much or as good data as they seem to think they have.)

It depends on what is your definition of reasonable. My definition of "reasonable" may well be different from yours. It depends on which toolboxes you own, as the optimization, curvefitting, and statistics toolboxes all have tools to make this task easier. You can also find useful tools on the file exchange.

In the end, it is impossible to know in advance (without some advance work, and looking at your data) whether that model can be fit easily to an unknown set of data.

[Joakim Ringblom]

Thank you for your answering, it was exactly what I wanted. Now I I´ll just get the right toolboxes and start trying.