Answers are grounded in this paper and cite their sources.
Predicting rare adverse events in clinical settings is complicated by extreme class imbalance, severe data scarcity, and the need for physiologically interpretable explanations. We propose Boundary-Stabilized SMOTE (BS-SMOTE), a domain-constrained oversampling method extending standard SMOTE with two constraints: feature-type aware generation applying distinct interpolation rules for continuous, binary, and categorical variables, and physiological feasibility projection clipping synthetic samples within clinically plausible bounds. To prevent data leakage, BS-SMOTE is applied strictly within each training fold of a stratified 5-fold nested cross-validation. The framework was applied to peri-anesthetic hypoxemia prediction in veterinary anesthesia using 138 dogs with 15 confirmed hypoxemic cases (imbalance ratio 1:8.2). A 10-way ablation study demonstrated that BS-SMOTE with fine-tuned GradientBoosting consistently outperformed standard SMOTE, CTGAN, and VAE-based augmentation (F1 = 0.307, AUC = 0.728, Cohen’s κ = 0.256). Robustness across 20 random seeds yielded stable discriminative performance (AUC = 0.751 ± 0.037). An augmentation ratio sweep from 0× to 8× confirmed monotonically improving class separability. SHAP analyses restricted to held-out original data identified respiratory compliance, age, and anesthesia duration as dominant risk contributors, with compliance exhibiting a nonlinear threshold effect consistent with veterinary respiratory physiology. BS-SMOTE provides a rigorous and interpretable framework for rare clinical event prediction, directly transferable to other data-scarce imbalanced settings.
No takes yet. Share an insight, caveat, or question.
Yu et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: