"Transformed Regression, Confusing Results"

Can somebody please explain how this operator works? Results seem confusing. After training a linear regression model inside this ensemble operator I applied the model. The resultant metadata contains an attribute named "prediction(transformation(label))", but when I try to use this attribute in "generate attribute" operator, It could not locate it. When I use prediction(label) instead, it is hard to explain but results seem unstable. Sometimes "generate attribute" uses just label attribute instead of prediction(label). Also, predictions take either 0 or 1 while the model is a linear regression.

An example is attached.