Skip to Content
0

Multivariate outlier detection in SAP PAL

May 19 at 06:17 AM

80

avatar image

Hi Experts,

We are slightly stuck with a question in terms of SAP PAL multivariate outlier detection and need help.

We are trying to implement a classification scenario with different variables( categorical + numerical ) but need to detect outliers in SAP PAL/ SAP BO PA tool first. We have around 300 records with 8 variables.

For single variable there are many algos in PAL like Grubb's , Inter quartile etc. but for multivariate I am not having much luck. Not able to find algos like MANOVA or Mahalanobis distance.

We cant do PCA here as we have categorical attributes too. We cant do individual column based outlier analysis as data is elliptical, so we need to do cumulative multivariate outlier analysis.

If possible can you please guide, as to how we can do this for multivariate scenario using PAL's algo ? We are really hoping if you can give us a direction, we would be highly obliged.

Thanks,

Hasan

10 |10000 characters needed characters left characters exceeded
* Please Login or Register to Answer, Follow or Comment.

2 Answers

Best Answer
Xingtian Shi
Jun 01 at 02:12 AM
0

If you use supervised learning algorithms where you have labelled outliers, you can use classification algorithms in PAL.If you don't have label information, you can use clustering algorithms like DBSCAN which return outliers information, or one-class SVM, which is newly available in PAL from HANA 2 SPS01.

Best regards,

Xingtian

Share
10 |10000 characters needed characters left characters exceeded
Hasan Rafiq May 20 at 03:58 AM
0

Dear experts,

Any guidance here?

One idea that's striking in my head is to use GMM or Hierarchical agglomerate clustering to do clustering and take small( smaller than ex: 5 ) clusters as outliers. However I am still not sure what is the best possibility in terms of SAP provided algos.

Really awaiting an expert advice.

Thanks for your support.

Hasan

Share
10 |10000 characters needed characters left characters exceeded