SubROC: AUC-Based Discovery of Exceptional Subgroup Performance for Binary Classifiers

TL;DR

SubROC is a framework that efficiently identifies subgroups where binary classifiers perform exceptionally well or poorly, aiding deployment decisions and model improvement.

Contribution

It introduces an open-source, interpretable, and efficient method for discovering subgroups with unusual classifier performance using AUC-based metrics.

Findings

Effective subgroup detection in diverse datasets

Improved search efficiency with pruning and significance testing

Practical utility demonstrated through case studies

Abstract

Machine learning (ML) is increasingly employed in real-world applications like medicine or economics, thus, potentially affecting large populations. However, ML models often do not perform homogeneously, leading to underperformance or, conversely, unusually high performance in certain subgroups (e.g., sex=female AND marital_status=married). Identifying such subgroups can support practical decisions on which subpopulation a model is safe to deploy or where more training data is required. However, an efficient and coherent framework for effective search is missing. Consequently, we introduce SubROC, an open-source, easy-to-use framework based on Exceptional Model Mining for reliably and efficiently finding strengths and weaknesses of classification models in the form of interpretable population subgroups. SubROC incorporates common evaluation measures (ROC and PR AUC), efficient search space pruning for fast exhaustive subgroup search, control for class imbalance, adjustment for redundant patterns, and significance testing. We illustrate the practical benefits of SubROC in case studies as well as in comparative analyses across multiple datasets.