eng
Schloss Dagstuhl – Leibniz-Zentrum für Informatik
Leibniz International Proceedings in Informatics
1868-8969
2023-02-01
92:1
92:20
10.4230/LIPIcs.ITCS.2023.92
article
Decision-Making Under Miscalibration
Rothblum, Guy N.
1
Yona, Gal
1
Weizmann Institute, Rehovot, Israel
How should we use ML-based predictions (e.g., risk of heart attack) to inform downstream binary classification decisions (e.g., undergoing a medical procedure)? When the risk estimates are perfectly calibrated, the answer is well understood: a classification problem’s cost structure induces an optimal treatment threshold j^⋆. In practice, however, predictors are often miscalibrated, and this can lead to harmful decisions. This raises a fundamental question: how should one use potentially miscalibrated predictions to inform binary decisions?
In this work, we study this question from the perspective of algorithmic fairness. Specifically, we focus on the impact of decisions on protected demographic subgroups, when we are only given a bound on the predictor’s anticipated degree of subgroup-miscalibration. We formalize a natural (distribution-free) solution concept for translating predictions into decisions: given anticipated miscalibration of α, we propose using the threshold j that minimizes the worst-case regret over all α-miscalibrated predictors, where the regret is the difference in clinical utility between using the threshold in question and using the optimal threshold in hindsight. We provide closed form expressions for j when miscalibration is measured using both expected and maximum calibration error which reveal that it indeed differs from j^⋆ (the optimal threshold under perfect calibration).
https://drops.dagstuhl.de/storage/00lipics/lipics-vol251-itcs2023/LIPIcs.ITCS.2023.92/LIPIcs.ITCS.2023.92.pdf
risk prediction
calibration
algorithmic fairness
multi-group fairness