End of Volume Review & Interview Questions

A card-fraud model that never flags a single transaction is 99.83% accurate. That is what a fraud rate of 0.172% does to a metric, and it is the rate in the public data that the payments company Worldline and the Machine Learning Group of the Université Libre de Bruxelles (ULB) released from two days of European card traffic in 2013: 492 frauds among 284,807 transactions, with most fields replaced by principal components to protect confidentiality. Its maintainers draw the first lesson themselves: measure the area under the precision–recall curve, because "confusion matrix accuracy is not meaningful for unbalanced classification."