ROC Curve is Flat with an Error about a Predictor containing values outside of 0 and 1

[kylegoodin]

I am trying to build a model for a PhD course that I am taking, and one of the requirements is that each type of model (Rich Forest, Decision Tree, etc) have an ROC Curve included. I believe I have everything set up correctly, but when I open the view for the curve, it is “flat”, with no curve to it. I also have a warning stating the following:
Prediction column “PEDS” contains values outside of range [0,1]. These values are clipped to [0,1].

While the PEDS column is included in my column filter, I also created a binned version of that column that returns “YES” if > 0 and “NO” if = 0 (essentially the column PEDS returns the number of pedestrians involved in an accident, and I only need “Yes/No”. Any advice would be greatly appreciated.

[k10shetty1]

Hi @kylegoodin,

Welcome to the forum.

In the prediction column of the ROC curve node, you need to pass the probability of the selected positive target class—not just 0 or 1 values. Supplying only binary prediction probabilities will result in a flat ROC curve.

Make sure your model outputs probability scores, and use those in the ROC Curve node to get a meaningful result.

Best,
Keerthan