Predicting which entity pairs are likely to break in the coming period, so the investigation starts before period end rather than after.
Classification over entity-pair features — historical break rate, transaction type, volume, time since last dispute
scikit-learn RandomForestClassifier