Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This study critically evaluates the common practice of assessing class-imbalance handling on single benchmark datasets, revealing that such conclusions can be misleading. Through a comprehensive analysis of 45 binary classification tasks with varying imbalance ratios, the authors demonstrate that threshold tuning significantly improves performance across most tasks, contrasting with findings from the Kaggle credit-card fraud dataset. Notably, the results indicate that the benefits of threshold tuning are non-monotonic and that calibration diagnostics do not reliably predict tuning advantages, challenging existing heuristics in the field.
Threshold tuning can enhance classification performance across diverse tasks, but its benefits are non-monotonic and often misrepresented when evaluated on single datasets.
Class-imbalance handling is routinely evaluated on a single benchmark dataset, and the resulting conclusions are reported as if they were properties of the method. We show this practice is unsafe. On the public Kaggle credit-card fraud dataset, under a leakage-free nested cross-validation protocol in which the decision threshold is selected on a held-out inner validation fold, a plain Random Forest at the default 0.5 threshold attains F1 = 0.861 +/- 0.021, and threshold tuning yields it no benefit (delta-F1 = -0.002). Read alone, this supports an appealing conclusion: for a well-calibrated ensemble, imbalance handling is unnecessary. We then apply the identical protocol to 45 binary tasks spanning imbalance ratios from 1:1.5 to 1:178 (2,025 model fits, four model families). The conclusion reverses. Random Forest benefits most from threshold tuning across the suite (delta-F1 = +0.101 +/- 0.134), not least, while three other families replicate their fraud-dataset behaviour almost exactly. SMOTE likewise harms the fraud dataset but helps across the suite (mean delta-F1 = +0.076; 138 wins, 39 losses; Wilcoxon p = 2.7e-17). Two further results. Threshold-tuning benefit is non-monotonic in the imbalance ratio: near zero below 1:5, peaking at +0.120 in the 1:15-1:40 band, declining to +0.045 beyond 1:100 - explaining why the fraud dataset, at 1:577, is an unrepresentative place to study the question. And we reject an intuitive heuristic: validation-set calibration error does not predict tuning benefit (expected calibration error r = -0.087; Brier r = +0.137), so calibration diagnostics cannot tell a practitioner whether tuning is worthwhile. We release the protocol, the 45-task harness, and all per-run metrics.