Identifies and addresses underrepresentation of groups based on gender, ethnicity, or other demographics in training data. By over-sampling from minority distributions, it creates balanced, synthetic datasets to train models that make equitable predictions, promoting algorithmic fairness.