LabVIEW

cancel
Showing results for 
Search instead for 
Did you mean: 

Need a method to determine weight for data points

I am trying to determine a best fit for a project of mine, however I do have some outling data points that affect the resulting best line fit.  It may not be a problem if I had many data points but I am working with only about 12-16 points to determine a non-linear best line fit. 

 

My stat class in college was rather limited and did not discuss any method for determing weighted values and I am struggling with coming up with an effective method myself.  Does anyone have a method, VI or link they could provide so I could read more upon the subject? 

 

Thanks,

 

Branden

0 Kudos
Message 1 of 4
(2,166 Views)
What kind of fit (linear, polynomial, nonlinear, etc.). Have you looked at the fitting palette? There are several options that deal well with outliers.
0 Kudos
Message 2 of 4
(2,153 Views)

I am using the nonlinear curve fit LM bound.  I did see the outliers VI but do not want to remove the data entirely, but want to reduce their affect on the best line fit.  These data points have a correlation to another function which is linear, and the outliers appear there as well.  I was going to use this correlation to determine a weight for the points used in the best line fit.

 

Branden 

0 Kudos
Message 3 of 4
(2,144 Views)

Try to play with the "method" input, e.g. bisquare.

 

A more detailed description is on the help page for linear fit or general polynomial fit, see the following image comparing how the methods deal with outliers.

 

0 Kudos
Message 4 of 4
(2,132 Views)