I want to apply a robust mahal distance and found an implementation in scikit: https://scikit-learn.org/stable/auto_examples/covariance/plot_mahalanobis_distances.html

but there is the number of outliers already given in advance. For me, who wants to find out the number of outliers, this makes no sense.. how can you know the number of outliers before applying that tool and what is its benefit then?

  • $\begingroup$ I am not sure about this but Mahalanobis distance seems to be used to find outlier in multivariate data. The resulting Mahalanobis distance is univariate and thus simple stats outlier methods (+- 3 standard deviation) or outside the interquartile range may be used. $\endgroup$ – BND Jul 26 '20 at 10:19
  • $\begingroup$ Its main purpose is to calculate the distance in multivariate data which, yes, can be used to determine outliers. You can choose the distance on your own (3, 5, 10 sds). $\endgroup$ – Ben Jul 26 '20 at 19:02

Your Answer

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

Browse other questions tagged or ask your own question.