Stacked conformal prediction

TL;DR

This paper introduces a method to conformalize stacked ensembles, enabling valid uncertainty quantification with manageable computation without separate calibration, and demonstrates its effectiveness empirically.

Contribution

It proposes a novel conformalization approach for stacked ensembles that is computationally efficient and achieves approximate marginal validity.

Findings

The method performs favorably compared to standard inductive conformal methods.

It maintains approximate marginal validity without a separate calibration sample.

Abstract

We consider a method for conformalizing a stacked ensemble of predictive models, showing that the potentially simple form of the meta-learner at the top of the stack enables a procedure with manageable computational cost that achieves approximate marginal validity without requiring the use of a separate calibration sample. Empirical results indicate that the method compares favorably to a standard inductive alternative.