System and method for deep interaction modeling for fraud detection

• US 20070090181A1
• Filed: 10/20/2005
• Published: 04/26/2007
• Est. Priority Date: 10/20/2005
• Status: Active Grant

• Current Assignee: Tech Mahindra Limited
• Original Assignee: Satyam Computer Services Limited (Tech Mahindra Limited)
• Inventors: Varadarajan, Sridhar, Gangadharpalli, Sridhar, Rao, Korrapati
• Granted Patent: US 7,552,865 B2
• Time in Patent Office: Days
• US Class Current: 235/380
• CPC Class Codes: G06Q 20/04 Payment circuits
• CPC Class Codes: G06Q 20/4016 involving fraud or risk level assessment in transaction processing
• CPC Class Codes: G06Q 20/403 Solvency checks
• CPC Class Codes: G06Q 40/00 Finance; Insurance; Tax strategies; Processing of corporate or income taxes

Abstract

The use of a variety of plastic cards to carry out monetary transactions is on the increase. It is important to ensure that such transactions are authenticated, authorized, and accounted in a best possible manner to prevent fraudulent transactions. However, as much the transactions are being protected, the fraudsters are becoming that much more cleverer and are being able to break into the system. One of the important requirements is to protect transactions of cash cards as these cards seem to be an easier target for fraudsters. A system for cash card fraud detection involves ensuring that a cash card user'"'"'s unique “signature” is determined and kept track of during the course of the use of the cash card.

Claims

1. A deep interaction modeling, DIM, system for fraud detection, said system comprising:

(a) Interaction Capture, IC, subsystem for capturing and storing user interactions comprising;

a Card Swipe operation Capture, CSC, element for capturing a plurality of user interactions during card swipe operation;

a Card Insertion operation Capture, CIC, element for capturing a plurality of user interactions during card insertion operation;

a PIN Insertion operation Capture, PIC, element for capturing a plurality of user interactions during PIN insertion operation;

a Screen Entry operation Capture, SEC, element for capturing a plurality of user interactions during screen entry operations;

a Workflow Navigation operation Capture, WNC, element for capturing a plurality of user interactions during workflow navigation operation;

a Withdraw Amount Capture, WAC, element for capturing a plurality of user interactions during amount withdraw operation; and

a Multi-Card Usage Capture, MCUC, element for capturing a plurality of user interactions during usage of a cash card;

(b) Interaction Analysis, IA, subsystem for analyzing and storing of captured user interactions comprising;

a Card Swipe Model, CSM, element for determining Card Swipe model weights;

a Card Insertion Model, CIM, element for determining Card Insertion model weights;

a PIN Insertion Model, PIM, element for determining PIN Insertion model weights;

a Screen Entry Model, SEM, element for determining Screen Entry model weights;

a Workflow Navigation Model, WNM, element for determining Workflow Navigation model weights;

a Withdraw Amount Model, WAM, element for determining Withdraw Amount model weights;

a Multi-Card Usage Model, MCUM, element for determining Multi-Card Usage model weights; and

a Fraud Assessment Model, FAM, element for determining Fraud Assessment model weights; and

(c) Interaction Verification, IV, subsystem for verifying captured user interactions comprising;

a Card Swipe Model based Fraudness Detection, CSMFD, element for determining fraudness of a transaction based on Card Swipe model;

a Card Insertion Model based Fraudness Detection, CIMFD, element for determining fraudness of a transaction based on Card Insertion model;

a PIN Insertion Model based Fraudness Detection, PIMFD, element for determining fraudness of a transaction based on PIN Insertion model;

a Screen Entry Model based Fraudness Detection, SEMFD, element for determining fraudness of a transaction based on Screen Entry model;

a Workflow Navigation Model based Fraudness Detection, WNMFD), element for determining fraudness of a transaction based on Workflow Navigation model;

a Withdraw Amount Model based Fraudness Detection, WAMFD, element for determining fraudness of a transaction based on Withdraw Amount model;

a Multi-Card Usage Model based Fraudness Detection, MCUMFD, element for determining fraudness of a transaction based on Multi-Card Usage model; and

a Fraud Assessment Model based Fraudness Detection, FAMFD, element for determining whether a transaction is fraudulent or not based on Fraud Assessment model.

2. The system of claim 1, wherein said CSC element comprises a procedure to capture a plurality of user interactions comprising:

false restart count, card swipe rate, and number of reswipes.

3. The system of claim 1, wherein said CIC element comprises a procedure to capture a plurality of user interactions comprising:

number of keystrokes before card insertion, false restart count, and number of keystrokes after card insertion.

4. The system of claim 1, wherein said PIC element comprises a procedure to capture a plurality of user interactions comprising:

false restart count, number of keystrokes before entering digit 1 of PIN, number of keystrokes before entering digit 2 of PIN, number of keystrokes before entering digit 3 of PIN, number of keystrokes before entering digit 4 of PIN, and number of keystrokes after entering PIN.

5. The system of claim 1, wherein said SEC element comprises a procedure to capture a plurality of user interactions, with respect to a screen comprising of a plurality of fields, comprising:

number of keystrokes before screen entry, sum of absolute difference between a field id of said plurality of fields and current position of said field id for each field id of said plurality of fields, and average inter-field interval.

6. The system of claim 1, wherein said WNC element comprises a procedure to capture a plurality of user interactions, with respect to a workflow comprising of a plurality of screens, comprising:

average inter-screen interval, and sum of absolute difference between screen id of a screen of said plurality of screens and current position of said screen for each screen of said plurality of screens.

7. The system of claim 1, wherein said WAC element comprises a procedure to capture a plurality of user interactions comprising:

number of keystrokes before providing amount, amount, a plurality of denominations, and number of keystrokes after providing amount.

8. The system of claim 1, wherein said MCUC element comprises a procedure to capture a plurality of user interactions during a withdraw transaction comprising:

card id of said withdraw transaction, amount of said withdraw transaction, hour of day of said withdraw transaction, and time of day of said withdraw transaction.

9. The system of claim 1, wherein said CSM element comprises a procedure to obtain a training data set, comprising:

false restart count, card swipe rate, number of reswipes, and fraudness score, related to Card Swipe model, and derive a neural network based Card Swipe model based on said training data set.

10. The system of claim 1, wherein said CIM element comprises a procedure to obtain a training data set, comprising:

number of keystrokes before card insertion, false restart count, number of keystrokes after card insertion, and fraudness score, related to Card Insertion model, and derive a neural network based Card Insertion model based on said training data set.

11. The system of claim 1, wherein said PIM element comprises a procedure to obtain a training data set, comprising:

false restart count, number of keystrokes before entering digit 1 of PIN, number of keystrokes before entering digit 2 of PIN, number of keystrokes before entering digit 3 of PIN, number of keystrokes before entering digit 4 of PIN, number of keystrokes after entering PIN, and fraudness score, related to PIN Insertion model, and derive a neural network based PIN Insertion model based on said training data set.

12. The system of claim 1, wherein said SEM element comprises a procedure to obtain a training data set, with respect to a screen comprising of a plurality of fields, comprising:

number of keystrokes before screen entry, sum of absolute difference between a field id of said plurality of fields and current position of said field id for each field id of said plurality of fields, average inter-field interval, and fraudness score, related to Screen Entry model, and derive a neural network based Screen Entry model based on said training data set.

13. The system of claim 1, wherein said WNM element comprises a procedure to obtain a training data set, with respect to a workflow comprising of a plurality of screens, comprising:

average inter-screen interval, absolute difference between screen id of a screen of said plurality of screens and current position of said screen for each screen of said plurality of screens, and fraudness score, related to Workflow Navigation model, and derive a neural network based Workflow Navigation model based on said training data set.

14. The system of claim 1, wherein said WAM element comprises a procedure to obtain a training data set, comprising:

number of keystrokes before providing amount, amount, a plurality of denominations, number of keystrokes after providing amount, and fraudness score, related to Withdraw Amount model, and derive a neural network based Withdraw Amount model based on said training data set.

15. The system of claim 1, wherein said MCUM element comprises a procedure to obtain a training data set, comprising:

card id, amount, hour, day, and fraudness score of a predefined number of cash card transactions, related to Multi-Card Usage model, and derive a neural network based Multi-Card Usage model based on said training data set.

16. The system of claim 1, wherein said FAM element comprises a procedure to obtain a training data set, comprising:

fraudness based on Card Swipe model, fraudness based on Card Insertion model, fraudness based on PIN Insertion model, Fraudness based on Screen Entry model, fraudness based on Workflow Navigation model, fraudness based on Withdraw Amount model, and fraudness based on Multi-Card Usage model, and fraud score, related to Fraud Assessment model, and derive a neural network based Fraud Assessment model based on said training data set.

17. The system of claim 1, wherein said CSMFD element comprises a procedure to verify said transaction for fraudness by computing a fraudness score based on Card Swipe model and comparing said fraudness score against a predefined threshold value.

18. The system of claim 1, wherein said CIMFD element comprises a procedure to verify said transaction for fraudness by computing a fraudness score based on Card Insertion model and comparing said fraudness score against a predefined threshold value.

19. The system of claim 1, wherein said PIMFD element comprises a procedure to verify said transaction for fraudness by computing a fraudness score based on PIN Insertion model and comparing said fraudness score against a predefined threshold value.

20. The system of claim 1, wherein said SEMFD element comprises a procedure to verify said transaction for fraudness by computing a fraudness score based on Screen Entry model and comparing said fraudness score against a predefined threshold value.

21. The system of claim 1, wherein said WNMFD element comprises a procedure to verify said transaction for fraudness by computing a fraudness score based on Workflow Navigation model and comparing said fraudness score against a predefined threshold value.

22. The system of claim 1, wherein said WAMFD element comprises a procedure to verify said transaction for fraudness by computing a fraudness score based on Withdraw Amount model and comparing said fraudness score against a predefined threshold value.

23. The system of claim 1, wherein said MCUMFD element comprises a procedure to verify said transaction for fraudness by computing a fraudness score based on Multi-Card Usage model and comparing said fraudness score against a predefined threshold value.

24. The system of claim 1, wherein said FAMFD element comprises a procedure to verify whether said transaction is fraudulent or not by computing a fraud score based on Fraud Assessment model and comparing said fraud score against a predefined threshold value.