Auditing machine learning algorithms

Overview

This paper discusses audits of machine learning (ML) algorithms by Supreme Audit Institutions (SAIs). The paper aims to help SAIs and individual auditors to perform audits on ML algorithms that have been applied by government agencies. It is designed for auditors with some knowledge of quantitative methods. Expert level knowledge of ML-models is not assumed.

We include an audit catalogue – a set of guidelines including suggested audit topics based on risks, as well as methodology to perform audit tests. The paper is accompanied by an Excel helper tool that sums up and guides through different parts of the audit.

External Links