Skip to main content

Table 1 The zero percentage TPR in the benchmark model is due to the fact that the static rule based model of the bank was not able to identify a single true positive case

From: Online payment fraud: from anomaly detection to risk management

Model

TPR

FPR

Benchmark model

0%

1% and 2%

500 LOF

18%

1%

500 LOF

45%

2%

500 BDT

18%

1%

500 BDT

27%

2%

500 IF

0%

1%

500 IF

27%

2%

Mean decision function

36%

1%

Mean decision function

36%

2%

Supervised decision function

18 %

1%

Supervised decision function

18%

2%

  1. The results of the mean aggregation approach derived from an aggregation of three standalone ensembles consisting of 500 LOF, 500 BDT and 500 IF base learners, where aggregation was performed by taking the mean value of decision functions from three different types of base learners