Bagging Provides Assumption-free Stability

TL;DR

This paper provides a distribution-free, finite-sample stability guarantee for bagging, demonstrating its effectiveness in stabilizing various models without assumptions on data or algorithm properties.

Contribution

It introduces a novel, assumption-free theoretical guarantee for bagging's stability applicable to many variants, validated by empirical results.

Findings

Bagging stabilizes highly unstable algorithms.

Guarantee holds without assumptions on data distribution.

Results are optimal up to a constant.

Abstract

Bagging is an important technique for stabilizing machine learning models. In this paper, we derive a finite-sample guarantee on the stability of bagging for any model. Our result places no assumptions on the distribution of the data, on the properties of the base algorithm, or on the dimensionality of the covariates. Our guarantee applies to many variants of bagging and is optimal up to a constant. Empirical results validate our findings, showing that bagging successfully stabilizes even highly unstable base algorithms.