Case repository

Year
Type of audit
Type of algorithm
Ethical issue
Harmonized standard
Algorithm owned by

Predicting irresponsible driving

Case study how irresponsible driving can be identified and predicted in the database of a car sharing platform. An independent commission issues advice on among others model validity, balancing false positives and false negatives and meaningful transparency. Read more...

TA:AA:2025:01 normative review machine learning profiling FP-FN balancing car sharing platform

Addendum Preventing prejudice

Further research into CUB process of Education Executive Agency of The Netherlands (DUO) by analysing aggregation statistics on the country of birth and country of origin of 300.000+ students in the period 2014-2022 provided by the Dutch national office of statistics Read more...

TA:AA:2024:02 technical audit rule-based profiling proxy discrimination DUO

Preventing prejudice

Disparities have been identified in the control process of a Dutch public sector organisation regarding misuse of college allowances. In the period 2012-2022, students who lived close to their parent(s) were significantly more often selected for a control procedure than others. The algorithm used to support the selection performed as expected. Read more...

TA:AA:2024:01 technical audit rule-based profiling proxy discrimination DUO

Risk Profiling for Social Welfare Reexamination

Case study to ML-driven risk predictions on unduly granted social welfare. An independent commission judges that algorithmic risk profiling can be used under strict conditions for sampling residents for re-examination. The aim of re-examination is a leading factor in judging profiling criteria. Read more...

ALGO:AA:2023:02 normative review profiling ML high-risk AI system proxy discrimination Municipality of Rotterdam

BERT-based disinformation classifier

Case study to algorithmic detection on fake news on Twitter. An independent advice commission believes there is a low risk of proxy discrimination by the BERT-based disinformation classifier and that the particular difference in treatment identified by the quantitative bias scan can be justified, if certain conditions apply. Read more...

ALGO:AA:2023:01 normative review bias detection tool ML high-risk AI FP-FN balancing disinformation

Type of SIM card as a predictor variable to detect payment fraud

Case study to ML-driven risk profiling to detect after-pay fraud at an e-commerce platform. An independent commission advises against using type of SIM card as an input variable in the algorithmic risk model. As it is likely that type of SIM card acts as a proxy-variable for sensitive demographic categories, the model would run an intolerable risk of disproportionally excluding vulnerable demographic groups from the payment service. Read more...

ALGO:AA:2022:01 normative review profiling proxy discrimination private organisation e-commerce

Newsletter

Stay up to date about our work by signing up for our newsletter

Newsletter

Stay up to date about our work by signing up for our newsletter

Building public knowledge for ethical algorithms