Automatic Generation of Attention Rules For Containment of Machine Learning Model Errors