Identification and management of fraudulent credit/debit card purchases at merchant ecommerce sites

• US 7,263,506 B2
• Filed: 02/12/2001
• Issued: 08/28/2007
• Est. Priority Date: 04/06/2000
• Status: Active Grant

• Current Assignee: Fair Isaac Corporation
• Original Assignee: Fair Isaac Corporation
• Inventors: Milana, Joseph P., Wilhelm, Wesley K., Lee, Walter W., Shao, Min
• Primary Examiner(s): Poinvil; Frantzy
• Assistant Examiner(s): Chencinski; Siegfried E.
• Application Number:
  US09/782,681
• Publication Number:
  US 20020099649A1
• Time in Patent Office: 2,388 Days
• Field of Search: 705/38, 705/44, 705/39, 705/76, 705/1
• US Class Current: 705/38
• CPC Class Codes: G06Q 20/04 Payment circuits
• CPC Class Codes: G06Q 20/10 specially adapted for electronic funds transfer [EFT] systems; specially adapted for home banking systems
• CPC Class Codes: G06Q 20/3821 Electronic credentials
• CPC Class Codes: G06Q 20/40 Authorisation, e.g. identification of payer or payee, verification of customer or shop credentials; Review and approval of payers, e.g. check credit lines or negative lists
• CPC Class Codes: G06Q 20/4014 Identity check for transactions
• CPC Class Codes: G06Q 20/4016 involving fraud or risk level assessment in transaction processing
• CPC Class Codes: G06Q 30/0185 Product, service or business identity fraud
• CPC Class Codes: G06Q 30/06 Buying, selling or leasing transactions
• CPC Class Codes: G06Q 40/03 Credit; Loans; Processing thereof
• CPC Class Codes: G06Q 40/06 Asset management; Financial planning or analysis

Abstract

Transaction processing of online transactions at merchant sites determines the likelihood that such transactions are fraudulent, accounting for unreliable fields of a transaction order, which fields do not reliably identify a purchaser. A scoring server using statistical model uses multiple profiles associated with key fields, along with weights to indicate the degree to which the profiles identify the purchaser of the transaction.

Claims

1. A computer implemented method for processing an online transaction, the method comprising the steps of:

generating from an online transaction of a purchaser, a plurality of keys from key fields of the transaction that individually do not reliably identify the purchaser;

for each key, retrieving a profile of historical transactions associated with the key;

weighting each profile with a weight indicating a degree to which the profile is associated with current purchaser;

generating a contrast measure using at least two of the profiles, and

generating a fraud score using said contrast measure indicative of the likelihood of fraud in the online transaction as a function of the weighted profiles and the current transaction.

2. The method of claim 1, wherein the key fields of the online transaction includes fields associated with any of purchaser identification information, order information, and payment information.

3. The method of claim 1, wherein each profile for a key includes a plurality of summary variables derived from the associated historical transactions, and which summarize the historical transactions having the key in a key field of the historical transactions.

4. The method of claim 1, further comprising the step of:

using the fraud score to determine whether to obtain additional information prior to completing the transaction.

5. The method of claim 4, further comprising the step of:

responsive to determining to obtain additional information, presenting a form with questions selected to obtain the additional information from the purchaser.

6. The method of claim 1, further comprising the step of:

using the fraud score to determine whether to hold the transaction for further review by a human analyst.

7. The method of claim 6, further comprising the step of:

responsive to determining to hold the transaction for further review, outsorting the transaction to a file for review by human analyst to determine whether to decline or approve the transaction, or obtain additional information prior to completing the transaction.

8. The method of claim 1, further comprising the step of:

using the fraud score to determine whether to approve the transaction or decline the transaction.

9. The method of claim 8, further comprising the step of:

responsive to determining to approve the transaction, completing the transaction order and fulfilling payment instructions for the order.

10. The method of claim 8, further comprising the step of:

responsive to determining to decline the transaction, transmitting a signal indicating that the transaction is declined.

11. The method of claim 8, wherein using the fraud score to determine whether to approve the transaction or decline the transaction further comprises the step of:

applying a plurality of rules to the fraud score and to the transaction.

12. The method of claim 8, wherein using the fraud score to determine whether to approve the transaction or decline the transaction further comprises the steps of:

storing velocity data based on an attribute of the transaction, the velocity data measuring a frequency of the attribute in a plurality of transactions; and

applying a velocity rule to the velocity data.

13. The method of claim 8, wherein using the fraud score to determine whether to approve the transaction or decline the transaction further comprises the step of:

declining the transaction if the fraud score is below a cutoff score, wherein the cutoff score is a function of a net margin of a merchant receiving the transaction.

14. The method of claim 8, wherein using the fraud score to determine whether to approve the transaction or decline the transaction further comprises the steps of:

for each of a plurality of different product categories, defining a different cutoff score; and

responsive to the product category pertaining to the transaction, applying the cutoff score for the product category to the fraud score.

15. A method for developing a statistical model of online transactions, the method comprising the steps of:

storing from a plurality of different online merchants, transactions for a plurality of different purchasers, each transaction indicated as being fraudulent or non-fraudulent;

for each of a plurality of key fields of the transaction, including at least one key field for which all of the possible keys of the key field do not reliably identify a purchaser of the transaction, generating for each key of the key field, a profile for all of the transactions having a matching key for the key field, such that at least one profile summarizes the transactions of a plurality of different individuals;

generating a contrast measure comparing the summary variables of at least two of the profiles; and

training a statistical model to generate a score for a transaction, the score indicative of a likelihood that the transaction is fraudulent, by using as training inputs to the statistical model selected transactions, the profiles associated with the keys of the key fields of each selected transaction, and the contrast measures associated with the profiles.

16. The method of claim 15, further comprising the steps of:

setting a cutoff score for rejecting a transaction by;

determining a desired transaction false positive rate as a function of a net margin;

selecting the score generated by the statistical model having an actual transaction false positive rate substantially similar or identical to the desired transaction false positive rate.

17. The method of claim 16, wherein determining a desired transaction false positive rate as a function of a net margin comprises the step of:

setting the transaction false positive rate TFPR according to the equation;

$\mathrm{TFRP}=\frac{1-\mathrm{Net}<br><br><br><br>\mathrm{Margin}}{\mathrm{Net}<br><br><br><br>\mathrm{Margin}}.$

18. The method of claim 15, wherein generating a contrast measure comprises computing a ratio of the summary variables of the at least two profiles.

38. The method of claim 15, wherein said step of training a statistical model uses a supervised learning environment.

39. The method of claim 15, wherein said model comprises any of:

a neural network; and

a regression model.

40. The method of claim 15, wherein said step of training a statistical model further comprises the step of:

preprocessing said at least one key field.

19. A computer implemented method for processing an online transaction, comprising the steps of:

generating from the online transaction of a purchaser, a plurality of keys from key fields of the transaction that individually do not reliably identify the purchaser;

for each key, retrieving a profile of historical transactions associated with the key, each profile including at least one summary variable;

computing at least one contrast measure for a summary variable included in the set of profiles; and

inputting the contrast measures, the online transaction data and the selected set of profiles into a predictive model to generate a fraud score indicative of the likelihood of fraud in the online transaction.

20. The method of claim 19, wherein the key fields of the online transaction includes fields associated with any of purchaser identification information, order information, and payment information.

21. The method of claim 19, wherein each profile for a key includes a plurality of summary variables derived from the associated historical transactions, and which summarize the historical transactions having the key in a key field of the historical transactions.

22. The method of claim 19, wherein computing at least one contrast measure for a summary variable comprises computing a ratio of the summary variables of the profiles.

23. The method of claim 19, further comprising the steps of:

comparing the fraud score to a plurality of thresholds to determine whether to approve the transaction, decline the transaction, obtain more information for the transaction, or hold the transaction for further review by a human analyst;

responsive to determining to approve the transaction, completing the transaction order, and fulfilling payment instructions for the online order, and completing the transaction;

responsive to determining to decline the transaction, transmitting to the purchaser/the merchant/a signal indicating that the transaction is declined;

responsive to determining to obtain additional information that is necessary to complete the transaction, presenting a form with questions selected to obtain the additional information from the purchaser; and

responsive to determining to hold the transaction for further review, outsorting the transaction to a file for review by human analyst to determine whether to decline, approve or obtain additional information.

24. The method of claim 23, wherein comparing the fraud score to a plurality of thresholds comprises the steps of:

storing velocity data based on an attribute of the transaction, the velocity data measuring a frequency of the attribute in a plurality of transactions; and

applying a velocity rule to the velocity data.

25. The method of claim 23, wherein comparing the fraud score to a plurality of thresholds further comprises the steps of:

for each of a plurality of different product categories, defining a different cutoff score; and

responsive to the product category pertaining to the transaction, applying the cutoff score for the product category to the fraud score.

26. The method of claim 25, wherein determining whether to decline the transaction further comprises the step of:

declining the transaction if the fraud score is below a cutoff score, wherein the cutoff score is a function of a net margin of a merchant receiving the transaction.

27. The method of claim 26, wherein the cutoff score as a function of a net margin comprises the step of:

setting the cutoff score to be a transaction false positive rate TFPR according to the equation;

$\mathrm{TFRP}=\frac{1-\mathrm{Net}<br><br><br><br>\mathrm{Margin}}{\mathrm{Net}<br><br><br><br>\mathrm{Margin}}.$

28. A computer implemented method for processing transactions to statistically identify a current purchaser, comprising the steps of:

receiving a transaction;

generating from the transaction a plurality of keys, including keys containing keys that individually do not reliably identify the purchaser;

for each key, retrieving a profile summarizing historical transactions by purchasers whose profiles match the key;

weighting each of the retrieved profiles by a weight indicating the degree to which the profile is associated with the current purchaser;

predicting the likelihood of fraud in the current transaction as a function of the weighted profiles and the current transaction.

29. A computer implemented method for identifying a current purchaser of online transaction, comprising the steps of:

receiving an online transaction from a current purchaser, the online transaction including purchaser identification information;

generating from the online transaction a plurality of keys that individually do not reliably identify the current purchaser;

for each key, retrieving a profile summarizing a plurality of historical transactions associated with the key, each profile including a plurality of summary variables derived from the associated historical transactions; and

selecting as the profiles of the current purchaser the profiles having the most similar summary variables.

41. The method of claim 29, wherein said summary variables describe historical purchasing behavior.

30. A system for processing online transactions, the system comprising:

a rule engine that receives from a scoring system a fraud score associated with a transaction, wherein said scoring system uses a summary variable in generation of said fraud score, the fraud score indicating the likelihood of fraud in the transaction, which applies a plurality of stored rules to the fraud score, each rule providing a condition and an action to perform in response to the transaction or the fraud score, to determine according to the rules whether to approve or decline the transaction, request more information from the purchaser, or hold the transaction for review by a human analyst;

an outsort management workstation that receives from the rule engine transactions to be held for review, stores the transactions in queues, and provides access to the queues to a human analyst in order to review transactions in the queues, the outsort management workstation further adapted to define for each queue at least one criteria for associating a transaction with the queue; and

a policy management workstation adapted to access the stored rules, and define rules for the rule engine to apply.

42. The system of claim 30, wherein said fraud score comprises:

a scaled score.

43. The system of claim 30, wherein said scoring system further uses a contrast measure in addition to said summary variable in generation of said fraud score.

31. A system for scoring a transaction, the system comprising:

storing means storing a plurality of stored profiles, each profile associated with a key for one of a plurality of key fields of a transaction, the key fields including at least one key field that does not reliably identify a purchaser of a transaction, each profile including summary variables summarizing all transactions having a same key for at least one of the key fields; and

a processor processing a statistical model that receives as inputs a transaction, a plurality of profiles, each profile summarizing transactions associated with a key for a key field, and at least one contrast measure that weights selected pairs of the profiles, and that produces a fraud score indicating the likelihood of fraud in the transaction.

32. The system of claim 31, wherein the at least one contrast measure is a ratio of the summary variables of a selected pair of profiles.

33. A method for establishing a cutoff score for a transaction processing system that processes transactions of a merchant, comprising the steps of:

providing a statistical model built in a supervised learning environment, wherein said statistical model generates a score categorizing a transaction, the score used by the merchant to accept or reject the transaction;

determining for each of a plurality of scores generated by the statistical model an actual transaction false positive rate;

determining a desired transaction false positive rate as a function of the merchant'"'"'s net margin; and

setting the cutoff score for rejecting transactions as a score having an actual transaction false positive rate approximating or equal to the desired transaction false positive rate.

34. The method of claim 33, wherein determining a desired transaction false positive rate as a function of the merchant'"'"'s net margin comprises the steps of:

setting the transaction false positive rate TFPR according to the equation;

$\mathrm{TFRP}=\frac{1-\mathrm{Net}<br><br><br><br>\mathrm{Margin}}{\mathrm{Net}<br><br><br><br>\mathrm{Margin}}$

wherein Net Margin is the merchant'"'"'s net margin.

45. The method of claim 33, further comprising the step of:

preprocessing data input into said model.

46. The method of claim 33, wherein said model is trained using a self-learning computerized system.

47. The method of claim 33, wherein said model comprises use of a neural network.

48. The method of claim 33, wherein said model comprises use of a contrast measure.

49. The method of claim 33, wherein said model comprises a regression model.

35. A computer implemented system for processing an online transaction, comprising:

means for receiving an online transaction from a current purchaser, the online transaction including purchaser identification information, order information, and payment information;

means for generating from the online transaction a plurality of keys that individually do not reliably identify the current purchaser;

means for retrieving a profile summarizing a plurality of historical transactions associated with the each key;

means for weighting each profile with a weight indicating a degree to which the profile is associated with current purchaser; and

means for generating a fraud score indicative of the likelihood of fraud in the online transaction as a function of the weighted profiles and the current transaction.

36. A system for developing a statistical model of online transactions, comprising:

a database storing transactions from a plurality of different online merchants, the transactions for a plurality of different purchasers, each transaction indicated as being fraudulent or non-fraudulent;

first generating means generating for each of a plurality of key fields of the transaction—

including at least one key field for which all of the possible keys of the key field do not reliably uniquely identify a purchaser of the transaction—

a profile for all of the transactions having a matching key for the key field, such that at least one profile summarizes the transactions of a plurality of different individuals;

second generating means for generating a contrast measure comparing the summary variables of at least two of the profiles; and

training means training a statistical model to generate a score for a transaction, the score indicative of a likelihood that the transaction is fraudulent, by using as training inputs to the statistical model selected transactions, the profiles associated with the keys of the key fields of each selected transaction, and the contrast measures associated with the profiles.

37. A system for processing online transactions, the system comprising:

a rule engine having a function of determining whether to approve a transaction, decline the transaction, request more information from the purchaser of the transaction, or hold the transaction for review by a human analyst by way of receiving a fraud score for the transaction from a scoring system and applying rules to the fraud score, wherein said scoring system uses a contrast measure in generation of said fraud score, wherein each rule defines a condition and an action to perform in response to the fraud score, to produce a determination for handling the transaction;

an outsort management workstation having a function of queuing transactions to be held for review and providing access to the queues to a human analyst in order to review transactions in the queues by way of receiving from the rule engine transactions to be held for review and storing the transactions in queues, each queue having at least one criteria for storing a transaction in the queue, to produce a set of queues, each queue storing one or more transactions; and

a policy management workstation having a function of defining rules for the rule engine to apply, by way of providing access to the stored rules.

44. The system of claim 37, wherein said scoring system comprises use of a summary variable in generation of said fraud score.