Diagnosing Model Performance Under Distribution Shift

TL;DR

This paper introduces DISDE, a method to decompose performance drops of prediction models under distribution shifts into components related to data difficulty, feature-outcome relationships, and unseen examples, aiding diagnosis and improvement.

Contribution

We develop DISDE, a novel approach that attributes performance degradation under distribution shifts to specific causes, enabling targeted model improvements.

Findings

DISDE effectively decomposes performance drops in real-world datasets.

The method provides insights into why domain adaptation methods may fail.

Application to census and satellite data demonstrates practical utility.

Abstract

Prediction models can perform poorly when deployed to target distributions different from the training distribution. To understand these operational failure modes, we develop a method, called DIstribution Shift DEcomposition (DISDE), to attribute a drop in performance to different types of distribution shifts. Our approach decomposes the performance drop into terms for 1) an increase in harder but frequently seen examples from training, 2) changes in the relationship between features and outcomes, and 3) poor performance on examples infrequent or unseen during training. These terms are defined by fixing a distribution on $X$ while varying the conditional distribution of $Y \mid X$ between training and target, or by fixing the conditional distribution of $Y \mid X$ while varying the distribution on $X$. In order to do this, we define a hypothetical distribution on $X$ consisting of values common in both training and target, over which it is easy to compare $Y \mid X$ and thus predictive performance. We estimate performance on this hypothetical distribution via reweighting methods. Empirically, we show how our method can 1) inform potential modeling improvements across distribution shifts for employment prediction on tabular census data, and 2) help to explain why certain domain adaptation methods fail to improve model performance for satellite image classification.