Bias and Fairness

Abstract

This chapter discusses sources of potential bias in the modeling pipeline, as well as some of the ways that bias introduced by a model can be measured, with a particular focus on classification problems. Interest in algorithmic fairness and bias has been growing recently, but it is easy to become lost in the large number of definitions and metrics. Related to the sample bias discussed in bias sources, understanding the relevant population for machine learning problem is important both to the modeling itself and to measures of equity. One mechanism by which bias may be introduced is in how the label/outcome itself is defined. For example, an analysis of recidivism might use a new arrest as an outcome variable when the appropriate outcome to look at may be the committing of a new crime. The chapter considers a set of more quantitative metrics that can be applied to the predictions of a machine learning pipeline specifically.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 23 citations exist upstream.