RUBRIC Improves Imbalanced Classification with Smart Oversampling

Class imbalance presents a significant hurdle in critical applications such as fraud detection and medical diagnosis, where minority-class samples are rare but crucial for accurate predictions. While oversampling methods generate synthetic data to balance class distributions, they often produce low-quality samples that can distort decision boundaries, lead to overfitting, and degrade generalization performance. Researchers introduce RUBRIC (Realism--Utility Balanced Ranking for Imbalanced Classification), a novel, generator-agnostic filtering framework designed to address these issues. RUBRIC reframes synthetic sample selection as a quality-over-quantity optimization problem. It ranks candidate synthetic samples by balancing two criteria: "realism," quantified by a discriminator distinguishing real from synthetic data, and "utility," which measures proximity to the decision boundary using a concave margin-based scoring function. The framework demonstrates that this filtering strategy, under mild conditions, tightens the generalization bound for margin-based classifiers by simultaneously reducing distribution shift and suppressing negative contributions from near-negative tail samples. Extensive experiments on credit-card fraud detection and other imbalanced benchmarks show that RUBRIC consistently improves F1-macro and recall while maintaining competitive ROC-AUC scores across various synthetic data generators.