Detect and treat local outliers

Mohamed El Hamrouni 231 Reputation points
2020-12-04T12:26:18.14+00:00

I have built a solution within Azure ML predicting the weight("Poids" below) of our materials production. It happens that some rows have a cost value that is not normal considering the similarity of the values from the other features. This row should be then exluded (replaced with a mean) according to me. How can I do this ?

** As this is not a real outlier in the all dataset but an outlier regarding values of the other fields.

** My predictive model gives the result of 19.000 where visually I would have expected 50.000 which is the correct value....

FT1Fini 20.00 20.00 20.00 20.00

FT2Fini 28.30 28.30 28.30 28.30

PagesTot 592 592 592 600

QteFact 27.120 29.045 29.045 27.973

PoidsKG 48.823 2.260 53.500 52.029

Thank you for your hep.

Regards,

Meddi.

Azure Machine Learning
Azure Machine Learning
An Azure machine learning service for building and deploying models.
2,563 questions
0 comments

1 answer

Sort by: Most helpful
Most helpful Newest Oldest
  1. GiftA-MSFT 11,151 Reputation points
    2020-12-04T18:44:23.903+00:00

    Thanks for reaching out. Try to use Clip Values data transformation to identify and optionally replace data values that are above or below a specified threshold with a mean, a constant, or other substitute value. Hope this helps.