When I replace my outliers with the median value of that column/feature, my mode for that column/feature also changes. Is that correct?
$\begingroup$
$\endgroup$
4
-
$\begingroup$ I suggest you windsorize or trim based on a quantile, e. g. 10%. However, it is important to investigate why these values were generated in the first place. $\endgroup$– 00schneiderCommented Feb 19, 2020 at 15:29
-
1$\begingroup$ @00schneider means winsorize. I agree that’s much, much better than replacing outliers with the median, but why replace them at all? $\endgroup$– Nick CoxCommented Feb 19, 2020 at 22:22
-
$\begingroup$ @00schneider thank you..so u mean to say that i should trim the top and bottom 10% of the tails? It is usually said that median is the best way to replace the outliers. How is trimming top and bottom 10% a better approach? or is it a special case with right skewed distributions?.. as in trimming is better than replacing with median when it comes to right skewed distributions? $\endgroup$– Sunaal DuaCommented Feb 20, 2020 at 5:55
-
$\begingroup$ I think you are confusing the treatment of outliers with the treatment of missing data. I recommend the literature of Wilcox for an introduction of the idea behind winsorizing/trimming. $\endgroup$– 00schneiderCommented Feb 20, 2020 at 7:25
Add a comment
|