Variable-Based Calibration for Machine Learning Classifiers

TL;DR

This paper introduces variable-based calibration to assess how well a model's confidence scores are calibrated across different data features, revealing limitations of traditional calibration metrics and proposing new detection and mitigation strategies.

Contribution

It generalizes calibration metrics to variable-based measures, demonstrating their importance and limitations, and offers methods for better calibration assessment and improvement.

Findings

Models with near-perfect ECE can still be miscalibrated across data features

Existing calibration methods may not address variable-based miscalibration

Variable-based calibration is crucial for fairness and interpretability

Abstract

The deployment of machine learning classifiers in high-stakes domains requires well-calibrated confidence scores for model predictions. In this paper we introduce the notion of variable-based calibration to characterize calibration properties of a model with respect to a variable of interest, generalizing traditional score-based metrics such as expected calibration error (ECE). In particular, we find that models with near-perfect ECE can exhibit significant miscalibration as a function of features of the data. We demonstrate this phenomenon both theoretically and in practice on multiple well-known datasets, and show that it can persist after the application of existing calibration methods. To mitigate this issue, we propose strategies for detection, visualization, and quantification of variable-based calibration error. We then examine the limitations of current score-based calibration methods and explore potential modifications. Finally, we discuss the implications of these findings, emphasizing that an understanding of calibration beyond simple aggregate measures is crucial for endeavors such as fairness and model interpretability.