FALCON-Discover: Discovering Concentrated False-Confidence Regions for Calibration
For practitioners using ML models in high-stakes settings, this work reframes calibration from aggregate metrics to localized failure detection, offering a practical tool to identify dangerous overconfidence regions.
The paper introduces FALCON-Discover, a post-hoc framework that identifies localized regions of high-confidence errors in predictions. Across seven tabular datasets, discrepancy-based ranking outperforms calibration baselines in recovering dangerous error mass, showing that false-confidence concentration is a recurrent but regime-dependent phenomenon.
Calibration is usually evaluated in aggregate, but the most dangerous failures are often local: predictions that remain highly confident despite being wrong. We study this failure mode as false-confidence concentration, the extent to which confident errors occupy compact, discoverable regions of prediction space. We introduce FALCON-Discover, a post-hoc, model-agnostic framework that ranks predictions using discrepancy signals from confidence, local support, neighborhood agreement, and perturbation stability. Across seven binary tabular datasets, four seeds, five-fold cross-fitting, and strong learners including XGBoost and CatBoost, we find that false-confidence concentration is recurrent but regime-dependent. At the main confidence threshold, discrepancy-based ranking substantially outperforms the strongest validation-selected calibration or trust-scoring baseline in the strongest regimes, while raw confidence recovers little dangerous-error mass. The best detector varies across datasets: learned discrepancy is strongest when multiple cues must be combined, whereas stability-centered ranking works best when local decisional fragility dominates. These results show that dangerous overconfidence is better treated as a family-level discovery problem than as a single-score calibration problem, and motivate calibration strategies that explicitly target regions where confidence, support, and stability diverge.