Detection of activity patterns

• US 10,395,309 B2
• Filed: 03/27/2008
• Issued: 08/27/2019
• Est. Priority Date: 03/30/2007
• Status: Active Grant

• Current Assignee: Symphonyai Netreveal Patent Limited
• Original Assignee: Detica Patent Ltd (BAE Systems Plc)
• Inventors: Kerley, Paul, O'Leary, Kevin Mary, Colleran, Kilian, Hayes, Cecil, Griffin, Liam
• Primary Examiner(s): Gurmu, Muluemebet
• Application Number:
  US12/450,528
• Publication Number:
  US 20100100693A1
• Time in Patent Office: 4,170 Days
• Field of Search: 707999006
• CPC Class Codes: G06Q 10/10 Office automation; Time management
• CPC Class Codes: G06Q 40/00 Finance; Insurance; Tax strategies; Processing of corporate or income taxes

Abstract

A monitoring system (1) comprises an interface (2) for receiving source alerts from at least one detection engine, a database (7) of historical events; and a classifier (3) for classifying received source alerts by linking a source alert with an historical event or a current source alert to provide a link, and providing said link as an output alert. The classifier comprises match methods (9) for processing source alerts and generating a score for extent of matching of a source alert with an historical event or current source alert, a voting engine (4) for weighting scores from the match methods (9), and a linking function (6) for determining that there is a link if a combination of the weighted outputs of a plurality of match methods exceeds a threshold. At least some match methods (9) are each associated with a specific field of a source alert such as a numerical value field or a name field of a source alert. A feedback function (6) notifies a case management system (5) of links, and the voting engine (4) receives from the case management system (5) feedback (11) of success of each match method (9), and adjusts match method weights (12) accordingly.

Claims

1. A fraud linking system for identifying links between determined fraud events detected in external systems, each external system having at least one fraud detection engine, the fraud linking system comprising:

an external interface configured to receive alerts reporting determined fraud events detected in the external systems by the respective at least one fraud detection engine;

a storage device configured to store determined historical fraud events in a database including the received determined fraud events in the alerts;

a processor comprising hardware configured to provide;

an alert classifier arranged to implement a plurality of match methods for detecting links between a received alert from one of the external systems and one or more stored determined historical fraud events or another received alert from another one of the external systems, each match method implementing different matching criteria, the alert classifier thereby generating and outputting a respective score for each match method representative of the extent of matching with one or more identified determined historical fraud events or another received alert; and

a voting engine for applying a set of weightings to the scores generated by the classifier for each of the match methods and for determining that there is a link with an identified determined historical fraud event or another received alert if a combination of the weighted scores exceeds a predetermined threshold.

2. The fraud linking system as claimed in claim 1, wherein the matching criteria of the plurality of matching methods comprise criteria defined in relation to specific field types in a received alert, selected from a numerical field, a name field and a risk quantifier.

3. The fraud linking system as claimed in claim 1, further comprising a case management sub-system arranged to receive notification of links determined by the voting engine and to receive feedback from a user regarding validity of a notified link, including an indication of whether the notified link comprises a false-positive or a false-negative linkage, and to forward the received feedback to the voting engine for use in adjusting the set of weightings.

4. The fraud linking system as claimed in claim 3, wherein the voting engine is arranged to adjust weightings in the set of weightings by applying linear or non-linear functions to feedback received from the case management sub-system to thereby favour scores generated in the alert classifier by match methods reliably indicative of valid links.

5. The fraud linking system as claimed in claim 4, wherein the voting engine is arranged to maintain a register of match method performance, using data stored in the register in the adjustment of weightings, and to update the register according to the received feedback, making use of register locking to ensure that updates due to feedback from more than one user may not be made simultaneously, thus preserving the integrity of the data stored in the register.

6. The fraud linking system as claimed in claim 3, further comprising a timer arranged periodically to trigger a process whereby a plurality of records of determined historical fraud events are extracted from the database according to predetermined search criteria and passed through the alert classifier and the voting engine to identify links between the extracted determined historical fraud events using the plurality of match methods or different match methods, any determined links being then recorded in the database using a unique key to identify the linked events as a group and notified to the case management sub-system.

7. The fraud linking system as claimed in claim 1, wherein the processor comprising hardware is further configured to provide an adapter for performing one or more functions upon receiving alerts, selected from:

normalizing a received alert;

alert validation;

addition of data, retrieved dynamically from an external database, to a received alert;

executing a set of logical routines on individual fields of a received alert to perform validation of fields of the alert to check that they satisfy quality tests;

executing a set of logical routines to perform transformation of individual fields in a received alert into a required format or create additional fields which are combinations of existing fields; and

routing a received alert directly to the alert classifier without modification.

8. The fraud linking system as claimed in claim 1, wherein the processor comprising hardware is further configured to provide an adapter for normalizing the alerts, and wherein:

the adapter performs alert validation, and

the adapter adds data to an alert, and

the adapter dynamically retrieves data from an external database to add to an alert, wherein the processor comprising hardware is further configured to route alert directly to the classifier, by-passing the adapter when a condition is satisfied; and

wherein the fraud linking system further comprises memory buffers to

temporarily hold messages until they can be processed.

9. The fraud linking system as claimed in claim 1, wherein:

the processor comprising hardware is further configured to provide an adapter for normalizing the alerts, and wherein the adapter performs alert validation, and

the adapter adds data to an alert, and

the adapter dynamically retrieves data from an external database to add to an alert, wherein the processor comprising hardware is further configured to route a alert directly to the alert classifier, by-passing the adapter when a condition is satisfied;

wherein the fraud linking system further comprises memory buffers to temporarily hold messages until they can be processed; and

wherein the processor comprising hardware is further configured to fetch messages from one of the memory buffers and to execute a set of logical routines on individual fields of each alert to perform validation of fields of the alert to check that they meet quality tests.

10. The fraud linking system as claimed in claim 1, wherein:

the processor comprising hardware is further configured to notify a case management system of links, and wherein

the voting engine comprises a function for receiving from the case management system feedback of success of each match method, and for adjusting match method weights accordingly,

the voting engine is adapted to initialize a plurality of match method weights to be equal and to vary them according to success of each individual match method in identifying links, and

said feedback includes a number of false positives and a number of false negatives; and

wherein said routines perform transformation of individual alert fields into a required format or create additional fields which are combinations of existing fields.

11. The fraud linking system as claimed in claim 1, further comprising a configuration function for configuring at least one of the components of the fraud linking system, including the alert classifier.

12. The fraud linking system as claimed in claim 1, further comprising a case management system for transmitting feedback to the alert classifier, and wherein:

the processor comprising hardware is further configured to notify a case management system of links,

the voting engine comprises a function for receiving from the case management system feedback of success of each match method, and for adjusting match method weights accordingly,

the voting engine is adapted to initialize a plurality of match method weights to be equal and to vary them according to success of each individual match method in identifying links, and

said feedback includes a number of false positives and a number of false negatives.

13. A non-transitory computer readable medium comprising software code for performing the alert classifier and the voting engine of a fraud linking system of claim 1 when the processor comprising hardware is a digital processor.

14. The fraud linking system as claimed in claim 6, wherein the processor comprising hardware is further configured to insert into the database a new database record comprising the unique key of the alert, the unique key(s) of one or more alerts and a flag to indicate that the alerts are linked.

15. The fraud linking system as claimed in claim 7, wherein the alert adapter translates received alerts so that they contain four fields including a unique identifier, name, date of birth and average transaction size.

16. The fraud linking system as claimed in claim 1, wherein the match methods include fuzzy text matching to find possible common sources based on descriptive attributes selected from name, address, occupation;

or quantitative attributes selected from Euclidean distance, Mahalanobis distance, transaction size, transaction frequency.

17. The fraud linking system as claimed in claim 1, wherein different received alerts indicate detected events associated with different persons.

18. The fraud linking system as claimed in claim 17, wherein the different received alerts indicate detected events associated with one of different persons and a same person, respectively.