1
$\begingroup$

which of these orders is correct?

First (Feature Selection) Second (Outlier Detection)

or

First (Outlier Detection) Second (Feature Selection)

$\endgroup$
  • $\begingroup$ Outlier detection in all your dimensional space -> feature selection. $\endgroup$ – user2974951 Oct 16 '18 at 10:06
  • $\begingroup$ @ user2974951, why did you select this order? $\endgroup$ – AHAD Oct 16 '18 at 10:09
  • $\begingroup$ In essence it depends on your objective / goal, however a data point may not be an outlier in a low dimensional space but it may be an outlier in a higher dimensional space, information that you would lose if you first performed variable selection. $\endgroup$ – user2974951 Oct 16 '18 at 10:12
1
$\begingroup$

In majority of the cases feature selection should be done after outlier detection. Outlier detection should be done at the initial stage of data pre-processing while feature extraction / selection can be don in the last part of data pre-processing.

Outlier must be detected beforehand so that actual behavior o that particular predictor is known. Then depending upon the behavior after removal of outlier it can be decided whether to keep that as feature or not.

Many feature selection algorithms like PCA, regression, etc are sensitive to outliers and so if such algorithms are used to extract the features then it would be better to remove the outliers beforehand.

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.