Can you both exclude outliers from a fit and use robust weighting for the remaining data?

In using the fit function, is it possible to simultaneously exclude outliers and use one of the robust fitting options to weight the remaining data? Is this ever warranted? I guess my question is partly having to do with the implementation and partly about what is appropriate or not conceptually. I have data where where the initial part fits to one distribution that I want to exclude, whereas the second part fits to the distribution that I want to fit.

Answer 1

[Image Analyst]

Why can't you just preprocess the data by removing outliers with rmoutliers() or other functions and then do the fitting? If there are only a few outliers, then they may not influence the fit that much. If there are lots of outliers, you can use something like RANSAC in the Computer Vision Toolbox.

[L'O.G.]

Thanks, I didn't know about that function. Is there any preferred reason to use that rather than both excluding outliers and using robust least squares simultaneously?

[Image Analyst]

RANSAC is normally used when there is a clear curve but it is buried in the presence of LOTS of noise. If you just have a little noise (like a few percent of points are "bad") then you should use isoutlier() or rmoutlier() or filloutlier().

Answer 2

[Sulaymon Eshkabilov]

In your exercise, if it is known which part to include in the fit simulation and which part to exclude, then you can use just appropriate indexes of your data for a fit model calculation.

If you want to remove just outliers from the data, then rmoutliers() can do the work easily.