[feature request] optionally provide a class prediction threshold in predictor nodes

KNIME Analytics Platform
#1

Hi,

Currently, class prediction / assignment is based on a fixed class probability threshold of 0.5, a not so practical value for real world classification applications. For now, class prediction can be changed a posteriori to the predictor node using Rule Engine. While this provides at least a working solution, the latter can be inconvenient due to the need to manually provide the possible values of the target variable, which is a source of error, and in addition unpractical when the target variable has more than two possible outcomes.

Cheers,

George

#2

Hi Geo,

I just opened a feature request in our bug tracing system.

However, it would mean adapting all of our predictor nodes. Is there any specific predictor you would need this for?

Best, Iris 

#3

Thank you, Iris!

This is a feature that will provide a huge convenience for any predictor node used for classification. Therefore, it is difficult to name a specific predictor.

Best, George