Replace extreme outliers in dataset

2 views (last 30 days)
Vlatko Milic
Vlatko Milic on 9 Apr 2019
Commented: Drewsky3 on 11 Dec 2019
Hi,
I want to replace "extreme" outliers in a data set. The data set varies quite a lot in terms of values, i.e. the maximum values are 50 times larger than the minimum values. However, in circa 0.05 % of the data points there are obvious mistakes depending on measurements errors. These values are at least 10 times larger than the maximum values in the data set. My question is therefore:
  • How can I replace the wrong value with the average of the value to the left and to the right of the wrong one? I know about the function "filloutliers" but the function finds values more than three local standard deviations away from the local mean. I need 5 or 6 means because of the large spread in data.
Kind regards!
  2 Comments
dpb
dpb on 9 Apr 2019
Post a typical dataset or two...
Vlatko Milic
Vlatko Milic on 10 Apr 2019
Hi, see attached file for two examples. In the 1st example I know there are some outliers and I suspect now that the original filloutliers that includes three local standard deviations away from the local mean might work. Do you have any advice maybe? :)
Thank you!

Sign in to comment.

Answers (2)

David Wilson
David Wilson on 10 Apr 2019
Have you considered hacking the source of filloutliers.m? If it is only a matter of scaling, then it is probably a simple one line change.
  1 Comment
Vlatko Milic
Vlatko Milic on 10 Apr 2019
Hi,
I did not know this was possible in Matlab. Is it difficult to hack the function? Any help is much appreciated :)
Thank you!

Sign in to comment.


David Wilson
David Wilson on 10 Apr 2019
Edited: David Wilson on 10 Apr 2019
Not difficult to hack, but slightly more complex to hack correctly. Just copy (and rename) the function into your work directory, and
>> which filloutliers
C:\Program Files\MATLAB\R2018b\toolbox\matlab\datafun\filloutliers.m
>> !copy "C:\Program Files\MATLAB\R2018b\toolbox\matlab\datafun\filloutliers.m" myFillOutliers.m
1 file(s) copied.
>> edit myFilloutliers
I'd start hacking on about line 335.
But there seems to be a problem calling the (invisible) function locateoutliers. You might have better luck carefully hacking the original function.
  3 Comments
dpb
dpb on 10 Apr 2019
I wouldn't recommend that route; bring the private functions need along as well so don't end up with base routine not doing what it does for other uses.
Plus, if do that and every install either a new release or a patch to the current, everything you've done will get overwritten and you'll have to do it all over again.
I used to do some things like that with a copule of particular functions that I didn't like their behavior but finally came to realize it was not the way to get the desired functionality for the longer term...
Drewsky3
Drewsky3 on 11 Dec 2019
Sooooo. . . . if you had an old lisence that never updated, and then someone had a new lisence, could you copy the new function library to the older version's program file?

Sign in to comment.

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!