Knime Target Shuffling

[bobrosski]

Hi,

I was wondering if someone might know why my Confusion Matrix in the Scorer Node is not showing any False Positives or True Positives when I use Target Shuffling. I only saw the Confusion Matrix show up accurately once out of the blue when using the Target Shuffling Node.

(Without Target Shuffling)

(With Target Shuffling)

[HansS]

Hi @bobrosski

I think of two possibilities you can look at.
Your have a very unbalanced data set regarding your target (99%=Y vs 1% N) or
there is a variable highly correlated with your target variable.

Hope this helps

gr
Hans

[beginner]

Yeah I would suspect the data set being unbalanced causing this. With target shuffling the data makes no sense anymore and the best the model can come up with is setting everything to the majority class.

[bobrosski]

Thanks for the feedback! I looked into wether the data is highly correlated or wether the dataset is unbalanced and this doesn’t seem to be the case:

Without Target Shuffling:

With Target Shuffling:

Without Target Shuffling:

With Target Shuffling:

Currently I’m trying to make predictions using:

Where tenure is a whole number > 0.

I noticed that when I add TotalCharges (a value > 0 with a maximum of two decimal places), the model that does not use target shuffling will sometimes display the False Positives and True Positives and other times it will not.