Table I. Performance scores on the test set when training the model on all transactions (A-D), excluding normal transactions (A) and excluding non-reported alerts/cases (B + C) in the training. The two rightmost columns contain performance scores on the test set with a multiresponse with three or four outcomes. All models are trained on 13,782 transactions. The Brier and AUC scores are evaluated on test sets with either only alert data (n_{test, only alerts} = 2,557) or all data (alert data and normal transactions, n_{test, all} = 4,967). The Brier and AUC scores are given with 90 per cent CI in brackets