For Qlik Predict to return a regression model instead of a binary one

[L_VN]

I have a training dataset of customers and if they accepted an offer; the target variable is a binary 1 or 0. I want to train a model that doesn't return a binary answer but rather a continuous score (in some range, maybe [0, 100). I do not seem to find an option for this in the settings of Qlik Predict.  

[igoralcantara · Accepted solution]

In order to have a regression, your target column on your training data needs more than 10 distinct values, otherwise is going to be treated as a classification.

What you can do it instead is to run a binary classification and use the positive class probability column, which you get once run a deployed model prediction.

That column is usually called “[TargetColumn_PositiveLabel]”. For example; Accept_1

[L_VN]

These are the results of a mock prediction. How do I interpret the Accept_0 and Accept_1? 

If Accept_0 close to 0 -> it predicts yes, and if Accept_1 close to 1 it predicts yes?

[igoralcantara]

I do not see any Accept_1 close to zero. This is a scale from 0 to 1. Multiply by 100 to get in %. All the 1 that I see are on the 99%, all the zeros are very close to zero, the biggest one is the first with 1.6%.

[L_VN]

The three last Accept_1 are of the order 10^-6, so I'd say that they are close to zero.

[igoralcantara]

Yes, that is why the prediction for them (column "Accept_Predicted") is 0. The probability of them being 1 is so low that the model predicted as 0.

[L_VN]

Ah, ok. I get it now. So for a classifier that can predict three classes (A, B, C) we'd get Accept_A, Accept_B, Accept_C?

[igoralcantara]

Exactly. To understand the cutoff point, check the training Experiment and look for the dot in the ROC chart.