Table 3 Feature importance.

From: Feature generation and contribution comparison for electronic fraud detection

Features

Feature categories

Importance

Rank

LT_Signature*

Others/non-fraudulent feature

1.80

1

AVG_T_Interval

Frequency/Basic statistical features

1.30

2

Second_Most_Frequent_Act

Frequency/Basic statistical features

1.07

3

Credit_STDev

Monetary/Basic statistical features

1.04

4

Saving_Gradient*

Others/Basic statistical features

0.97

5

Most_Frequent_Act

Frequency/Basic statistical features

0.78

6

Debit_STDev

Monetary/Basic statistical features

0.76

7

24hr_Trade

Recency/Basic statistical features

0.73

8

Note_Not_NE

Others/Basic statistical features

0.70

9

BankID

Others/Basic statistical features

0.69

10

Least_Frequent_Act

Frequency/Basic statistical features

0.68

11

Magic_Number_Check

Monetary/fraudulent feature

0.57

12

Weighted_Fraud_Coefficient*

Others/fraudulent feature

0.53

13

LDT >R*

Others/fraudulent feature

0.50

14

Withdrawal_After_Deposit*

Others/fraudulent feature

0.48

15

MFTAC*

others/fraudulent feature

0.46

16

LATM_Signature*

Others/non-fraudulent feature

0.27

17

HFW_Bank_of_ATM*

Others/fraudulent feature

0.20

18

Suspicious_Branch

Others/fraudulent feature

0.17

19

  1. Features marked with “*” are unique features in our model and features listed as “Other” are not RFM or FBADT features.
  2. (FBADT features do not appear here because the scores are too low and they are not included in Other).