Suspicion Machines Methodology

Summary

Detailed methodology document describing fairness testing using statistical parity and controlled statistical parity measures. Reports women 1.25x more likely to be flagged, non-Dutch speakers almost 2x more likely, parents 1.7x more likely. Age was the most important variable with nearly 3x the impact of the second-ranked variable. Training data contained 50% fraud cases versus 21% actual population rate. ROC analysis showed system only 50% more accurate than random. Netherlands Institute of Human Rights concluded language variable constituted indirect discrimination on basis of origin.