Ensemble Threshold Calibration for Stable Sensitivity Control
This addresses the critical need for stable sensitivity control in large-scale matching applications, where missing true matches can break downstream analytics and excessive manual review inflates costs, representing a strong incremental improvement over classical calibration methods.
The paper tackles the problem of achieving precise recall control in large-scale spatial conflation and entity-matching tasks, where existing methods like Clopper-Pearson or Wilson often overshoot targets with high variance. The result is an end-to-end framework that achieves exact recall with sub-percent variance on datasets of up to 67.34 million pairs while reducing redundant verifications and running efficiently on a single TPU core.
Precise recall control is critical in large-scale spatial conflation and entity-matching tasks, where missing even a few true matches can break downstream analytics, while excessive manual review inflates cost. Classical confidence-interval cuts such as Clopper-Pearson or Wilson provide lower bounds on recall, but they routinely overshoot the target by several percentage points and exhibit high run-to-run variance under skewed score distributions. We present an end-to-end framework that achieves exact recall with sub-percent variance over tens of millions of geometry pairs, while remaining TPU-friendly. Our pipeline starts with an equigrid bounding-box filter and compressed sparse row (CSR) candidate representation, reducing pair enumeration by two orders of magnitude. A deterministic xxHash bootstrap sample trains a lightweight neural ranker; its scores are propagated to all remaining pairs via a single forward pass and used to construct a reproducible, score-decile-stratified calibration set. Four complementary threshold estimators - Clopper-Pearson, Jeffreys, Wilson, and an exact quantile - are aggregated via inverse-variance weighting, then fused across nine independent subsamples. This ensemble reduces threshold variance compared to any single method. Evaluated on two real cadastral datasets (approximately 6.31M and 67.34M pairs), our approach consistently hits a recall target within a small error, decreases redundant verifications relative to other calibrations, and runs end-to-end on a single TPU v3 core.