Curve to fit a loglog graph and its equation

i have write a code for my given set of data and could not figure how to get a curve line to fit it with an equation

this is the code i have writen with the data points

y=[0.02521 0.47575 0.035403 0.03479 0.025683 0.025397 0.021046];

x=[0 0.5 1 1.5 2 2.5 3];

y=y./1000;

loglog(x.',y.','o')

hold on

f=polyfit(log(x),log(y),1);

fity=polyval(f,log(x));

loglog(x,exp(fity))

[Sam Chak]

There are only 7 points, so probably insufficient to produce a good fitting. Try to obtain more data points.

This is just one of many possible equations to fit the 7 points:

Goodness of fit:

• SSE: 0.0001094
• R-square: 0.9994
• Adjusted R-square: 0.9992
• RMSE: 0.004677

[Uygur Kinay]

I have 30 data for the equation and I got two equations in excel one is polynominal and other is logoritmic equation.

How we can do it on matlab,Could you guide me?

Answer 1

[Thiago Henrique Gomes Lobato]

There are some issues with your fit. First, log(0) is not defined, so what you expect to become from a fit that uses log(0)? If you really want to use this point to a logarithmic fit you have to make some consideration like calculating log(x+eps) instead of log(x), although this will unlikely lead to a reasonable model. Second, what type of function you think you have? The second y value has a way higher value than all others, it is an outlier/measurement error that needs to be removed? Or you are modelling a probability function and there is the peak? If so, why use a 1 degree exponential function and not a probability function such as a gaussian to model it?

If you don't define what type of model you're expecting to have it doesn't make much sense to make a fit on it.