What do you want to fit? A general trend or the most significant of the oscillations?
The polyfit function will do best for a general trend.
For the oscillations, I would start with a fft of it to find one or more of the most significant frequencies. Then use one of the nonlinear curve fitting functions with an objective function:
fs = @(b,x) b(1) + b(2).*sin(b(3).*x + b(4)) + b(5).*sin(b(6).*x + b(7));
I would not use more terms or parameters. Two terms is probably the most you can hope to fit successfully, and that may be optimistic.