How to define outlier by 𝑀𝑒𝑎𝑛 ±𝑆𝐷

[Gambit]

Hi Guys,

Need your help. Currently I am removing outliers by Mean ± (factor)𝑆𝐷.
I am using outlier removal node but it can only calculate integer types.
Do you know other solutions?

Thanks in advance Guys.

Regards,
Gambit

[ScottF]

Hi @Gambit -

The Outlier Removal node is an older community node. Maybe you could try the Numeric Outliers node if the IQR method is an option for you?

Alternately, you could always calculate your cutoff using mean and SD manually using a Math Formula node and then apply the cutoff with a subsequent Row Filter.

There are several workflows on the KNIME Hub showcasing various methods of outlier detection as well.

[Gambit]

Hi @ScottF,

Thank your for your response. IQR method is also one of my options and I already used it on my other workflows. Currently, I need to define the outliers using Mean ± (factor)𝑆𝐷. I am new with Analysis Based on my understanding, Standard Deviation can be calculated from a column, same goes with Mean. But can you guide me how to identify the outliers on a single column using this method? I apologize for the inconvenience. Your advice would help me a lot!

Thanks and regards,
Gambit

[ScottF]

Try this example:

Manual_Outlier_Calculation_Example.knwf (14.7 KB)

Note that it uses flow variables, in case you aren’t familiar with them. But other than that I think it’s self-explanatory.

[Daniel_Weikert]

You specified the bounds in the Double Cell variable in Row Filter. What would be a use case for datacell there (it’s a string?) Maybe there is a documentation for this? Thanks

[ScottF]

That’s a good question, and my answer at the moment is a very unsatisfying “I don’t know”.

There’s not documentation for the Row Filter flow variables specifically that I know of. Based on the error message returned when I try to supply a flow variable to the datacell field, I’m wondering if that’s supposed to be exposed in the flow variables dialog at all.

[Gambit]

Hi @ScottF

Thank you for this! It really helped a lot!

Best regards,
Gambit