My Scorecard has null values - what happened?

Mahshid
Mahshid
Altair Employee
edited May 2021 in Community Q&A

There's a few reasons why your scorecard could end up with null values, all of which can be 'solved' by re-configuring the WOE node or Scorecard node.

Reason 1: You have values that aren't accounted for in your training dataset, this might include new categories, or values which are beyond the range of the training dataset values (e.g. you have a 120 year old person, and the oldest person was 100 years in your training dataset).

Fix 1: Depending on whether this is something that you want to highlight, it can be resolved in the scorecard node, by changing 'Score for Outliers'. It could be good practice to leave this as null, as these tend to be unusual circumstances.

image



Fix 2: For continuous variables, in the WOE node you can select 'Attribute Editor' to modify the variables...

 

image


...and then allow 'Open left' and 'Open right', which will set the limits to +/- infinity.

image


Note: you can shift/ctrl select to change multiple rows at a time

Reason 2: You might have NULL values and aren't assigning them to their own category in your WOE this will end up not providing a score for NULL.

Fix: Again, go to the WOE node attribute editor. Now change the 'Missing values' field to 'Use'. This creates a new bin in the decision tree/WOE transform, specifically for NULL values.

image

Reason 3: You haven't enforced 'no pure nodes' or a minimum number of each side of the dependent variable as > 0, because the WOE is created based on a ratio of each side of the dependent variable - e.g. Good:Bad it will have a null value if there are no Bads.

Fix: In the WOE node, apply the constraint 'No pure nodes':

image

 

Or do this globally with the 'Optimize All' option at the top of the screen:


image



------------------------------
Alex Gobolos
Sales Engineer
Datawatch Corporation
Toronto ON
------------------------------