Hybrid approaches address such trade-offs by combining oversampling with post-resampling cleaning strategies that remove noisy or duplicate samples. A classic example is SMOTE combined with Edited Nearest Neighbors (SMOTE+ENN), which, after oversampling minority-class samples through synthetic generation techniques, performs neighbourhood cleaning to remove uncertain samples, thereby increasing both class separability and instability \cite{Fernandez18}. Hybrid approaches are typical for biomedical machine learning, as high-throughput data is often noisy, and the value of a feature often depends on other features

Hybrid methods tackle these compromises by merging oversampling techniques with cleaning processes that eliminate noisy or redundant samples after resampling. The SMOTE combined with Edited Nearest Neighbors approach (SMOTE+ENN) serves as a well-known illustration of this concept. This method first generates synthetic samples to increase minority-class representation, then applies neighborhood-based cleaning to eliminate ambiguous samples, which enhances class separability while introducing some instability \cite{Fernandez18}. These hybrid strategies are commonly used in biomedical machine learning applications because high-throughput datasets frequently contain noise, and individual feature values typically rely on interactions with other features.