Training technique that uses alternative rule outcomes to improve model accuracy in applying rule effects to objects and attributes.