Instance-Level Costs for Nuanced Classifier Evaluation

Kabir Kang, Stephen Mussmann

Standard classification treats all errors equally, but in applications such as content moderation and medical screening, mistakes on clear-cut cases are more costly than errors on…