Worst Group Accuracy

Worst-group accuracy (WGA) focuses on improving the performance of machine learning models on the least-represented subgroups within a dataset, addressing the issue of unfair or biased predictions. Current research emphasizes mitigating spurious correlations—where models exploit irrelevant features—through techniques like last-layer retraining, data augmentation (e.g., reweighting, upsampling), and multitask learning, often incorporating pre-trained models and refined group inference methods. Improving WGA is crucial for ensuring fairness and reliability in machine learning applications, impacting diverse fields from healthcare to criminal justice by promoting equitable outcomes across all populations.