The SMOTE Paradox: Why a 92% Baseline Collapsed to 6%—A Systematic Review of 821 Papers in Imbalanced Learning (2020–2025)

TMLR Paper6827 Authors

Abstract:

Class imbalance pervades production systems—fraud detection, medical diagnosis,
industrial monitoring—yet handling it effectively remains challenging. For two
decades, SMOTE has been the default solution, but practitioners increasingly
abandon it at scale.

We investigate this disconnect through systematic review of 821 DBLP papers
(2020–2025) and bibliometric analysis of 4,985 Scopus records. Our analysis
reveals the SMOTE Paradox: only 6% of high-impact papers successfully executed
SMOTE at full scale due to memory exhaustion or preprocessing bottlenecks.
The field has fragmented, with 30% adopting generative models, 30% using
cost-sensitive losses, and 40% employing hybrid approaches.

Three factors explain SMOTE's decline. First, $O(N \cdot N_{\text{min}} \cdot d)$
nearest-neighbor search requires 1.28 TB memory for typical modern datasets.
Second, linear interpolation produces off-manifold artifacts scaling as $\sqrt{d}$
in high dimensions. Third, CPU-bound preprocessing creates friction with
GPU-centric training pipelines.

We validate these findings through controlled experiments across seven datasets
(196 trials, imbalance ratios 1.1:1 to 129:1). Statistical testing reveals
no significant ROC-AUC differences between SMOTE and cost-sensitive baselines
(Friedman $p=0.907$), despite SMOTE incurring 2.7× computational overhead.
However, cost-sensitive methods severely degrade at extreme imbalance (>40:1).

Submission Type: Long submission (more than 12 pages of main content)

Assigned Action Editor: ~Ju_Sun1

Submission Number: 6827