Suppose I have a dataset of 20 features, X1, X2..X20.

Say I perform an outlier detection algorithm such as One-class SVM (http://scikit-learn.org/stable/modules/generated/sklearn.svm.OneClassSVM.html) or IsolationForest (http://scikit-learn.org/stable/modules/generated/sklearn.ensemble.IsolationForest.html) on the dataset.

Suppose that I have some domain knowledge, such as if X1<10 the data point is really abnormal, or if X2==1 the value of X4 is not important anymore etc.

How could I integrate these information to One-class SVM or IsolationForest?

  • $\begingroup$ As an independent check of the validity of the outlier detection rule? $\endgroup$ – user603 Aug 20 '18 at 10:39
  • $\begingroup$ Could you elaborate a little bit more? $\endgroup$ – mommomonthewind Aug 20 '18 at 10:40
  • $\begingroup$ You have two outlier detection algorithm that you see fit to use. Clearly they will not give you the same list of outliers. Which algorithm should you use? Maybe the one that overlaps the most with the expert's list. At list you will have the experts on your side! $\endgroup$ – user603 Aug 20 '18 at 10:41
  • $\begingroup$ But the expert knowledge is not complete - it is kind of guidance rather than a complete set of rules. $\endgroup$ – mommomonthewind Aug 20 '18 at 11:09
  • $\begingroup$ sure, as long as the expert guidance are is not wrong this strategy will help. $\endgroup$ – user603 Aug 20 '18 at 11:16

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.