Reviewing Possibly Mislabeled Pairs

This suggestion is for reviewing potentially mislabeled pairs. It is based on the distance (absolute difference) between the actual label and the score provided by the trained model. The actual label represents score 0.0 or 1.0 for “False” and “True” labels respectively. Record pairs where the desired score for the actual label and the predicted score are very distant are included in the suggestion.