Performance Estimation in Binary Classification Using Calibrated Confidence
Juhani Kivimäki, Jakub Białek, Wojtek Kuberski, Jukka K. Nurminen
TL;DR
CBPE presents a framework for unsupervised performance estimation of binary classifiers across multiple confusion-matrix metrics by leveraging calibrated confidence scores. It treats confusion-matrix elements as random variables to derive the full distribution—and thus confidence intervals—for $Accuracy$, $Precision$, $Recall$, and $F_1$, while offering fast point-estimate shortcuts. The method provides coverage guarantees under perfect calibration and demonstrates practical effectiveness via synthetic experiments and the TableShift benchmark, highlighting strengths and limitations under distribution shifts. This work enables uncertainty-aware monitoring of deployed classifiers without access to ground-truth labels, supporting better detection of performance changes in real-world settings with class imbalance and varying costs.
Abstract
Model monitoring is a critical component of the machine learning lifecycle, safeguarding against undetected drops in the model's performance after deployment. Traditionally, performance monitoring has required access to ground truth labels, which are not always readily available. This can result in unacceptable latency or render performance monitoring altogether impossible. Recently, methods designed to estimate the accuracy of classifier models without access to labels have shown promising results. However, there are various other metrics that might be more suitable for assessing model performance in many cases. Until now, none of these important metrics has received similar interest from the scientific community. In this work, we address this gap by presenting CBPE, a novel method that can estimate any binary classification metric defined using the confusion matrix. In particular, we choose four metrics from this large family: accuracy, precision, recall, and F$_1$, to demonstrate our method. CBPE treats the elements of the confusion matrix as random variables and leverages calibrated confidence scores of the model to estimate their distributions. The desired metric is then also treated as a random variable, whose full probability distribution can be derived from the estimated confusion matrix. CBPE is shown to produce estimates that come with strong theoretical guarantees and valid confidence intervals.