Researchers say they have developed a simple model for buyers to predict with more than 80% accuracy when an order is likely to be sent to an unapproved factory.