Identity verification methods

• US 6,430,305 B1
• Filed: 12/20/1996
• Issued: 08/06/2002
• Est. Priority Date: 12/20/1996
• Status: Expired due to Term

• Current Assignee: Synaptics Incorporated
• Original Assignee: Synaptics Incorporated
• Inventors: Decker, Joseph E.
• Primary Examiner(s): Johnson, Timothy M.
• Application Number:
  US08/771,707
• Time in Patent Office: 2,055 Days
• Field of Search: 382/119, 382/156, 382/157, 382/115, 382/116, 382/155, 382/159, 382/160, 382/186, 382/187, 382/224, 382/226, 382/228, 382/137, 382/138, 382/135, 382/124, 235/379, 235/380, 235/382, 235/382.5, 902/2, 902/3, 902/24, 902/5, 902/25, 705/1, 705/35, 705/38, 705/39, 705/44, 348/161, 706/15, 706/20, 340/825.3, 340/5.2, 340/5.21, 340/5.3, 340/5.4, 340/5.41, 340/5.53, 340/5.6, 340/5.65, 340/5.66, 340/5.74, 340/5.8, 340/5.81, 340/5.83, 380/23-25
• US Class Current: 382/116
• CPC Class Codes: G06F 18/254 of classification results, e.g. of results related to same input data
• 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

Abstract

Apparatus for fraud detection for an account includes a plurality of N statistical estimators which estimate the likelihood of a transaction being fraudulent from data about a transaction on the account and transaction history of the account. A statistical estimator is provided which computes an estimate of the likelihood of a particular signature being that of a person authorized to sign on the account from an exemplar of the signature and a history of previous signatures. A combiner produces a combined probability estimate from the plurality of N statistical estimators and the statistical estimator.

Claims

1. An apparatus for fraud detection for an account on the event of a transaction attempted on the account, comprising:

an electronic signature capture device operatively coupled to a signature processing system;

a data collection device;

a database operatively coupled to said signature capture device and said data collection device;

a plurality of statistical estimators operatively coupled to said database which estimate the likelihood of a transaction being fraudulent;

a statistical signature validity estimator operatively coupled to said database which computes the likelihood of a particular signature from said signature processing system being that of a person authorized to sign on the account from an exemplar of said particular signature and a history of previous signatures, said statistical validity estimator performing estimation independently of said plurality of statistical estimators; and

a combiner for producing a combined probability estimate from said plurality of statistical estimators and said statistical signature validity estimator.

2. The apparatus of claim 1 wherein said combiner is a neural network.

3. The apparatus of claim 1 wherein said combiner is a tree-based regressor.

4. The apparatus of claim 1 wherein said history of previous signatures have a probability of having been made by a person other than said person authorized to sign on the account which is less than some threshold probability.

5. The apparatus of claim 1 wherein said account is a credit card account which an authorized user is provided with a physical card capable of storing information and wherein information representing said history of previous signatures is stored thereon and wherein said statistical signature validity estimator is located at a location where a person is attempting said transaction.

6. The apparatus of claim 1, wherein said plurality of statistical estimators and said statistical signature validity estimator are selectively updateable.

8. The method of claim 6, further comprising generating a cost of accepting the transaction by weighing said combined probability estimate against a cost of denying said transaction.

7. A method for fraud detection for a transaction attempted on an account including the steps of:

providing a plurality of statistical estimators which estimates the likelihood of the transaction being fraudulent from data about a transaction on the account and transaction history of the account;

providing said plurality of statistical estimators with data about a transaction on the account and transaction history of the account;

providing computer data representing a signature from an electronic signature capture device;

providing a selectively updateable statistical estimator which computes an estimate of the likelihood of a particular signature being that of a person authorized to sign on the account from an exemplar of the signature and the history of previous signatures, said statistical estimator performing estimation independently of said plurality of statistical estimators;

.

providing said statistical estimator with data comprising an exemplar of the signature and a history of previous signatures; and

producing a combined probability estimate from the plurality of statistical estimators and the statistical estimator.

9. A method for fraud-pattern detection, including the steps of:

providing a multi-layer neural network combiner having N “

zero-one”

inputs and one “

zero-one”

output;

collecting training data for said combiner, including the steps of recording a plurality of signature verification attempts and their associated transaction, and keeping track of the transactions which are fraudulent transactions;

training said multi-layer neural network combiner on said N “

zero-one”

inputs and one “

zero-one”

output, wherein the N inputs are used to represent the probabilities output from each of the individual estimators which are to be combined using on-line backpropagation with targets of zero for those transactions which turned out to be fraudulent and one for those transactions which turned out to be valid; and

evaluating individual transactions, by first generating the individual probabilities, and combining them by presenting the individual probabilities as inputs to the neural network.

10. A method for fraud detection in an electronic transaction by a purchaser, including the steps of:

providing at a buyer'"'"'s computer an account number for a particular credit card account;

creating at said buyers computer data representing a signature from an electronic signature capture device;

transmitting said account number and said data representing a signature from said buyer to a seller over a computer network;

transmitting said account number and said data representing a signature from said seller to a credit card issuer over a computer network;

providing said credit card issuer a plurality of statistical estimators which estimate the likelihood of the transaction being fraudulent from data about a transaction on the account and transaction history of the account;

providing said credit card issuer a statistical estimator which computes an estimate of the likelihood of a particular signature being that of a person authorized to sign on the account from an exemplar of the signature and a history of previous signatures, said statistical estimator performing estimation independently of said plurality of statistical estimators;

providing said credit card issuer said statistical estimator with data comprising an exemplar of the signature and the history of previous signatures;

producing at said credit card issuer a combined probability estimate from plurality of statistical estimators and the statistical estimator;

transmitting an acceptance or non-acceptance of the transaction from said credit card issuer to said seller; and

providing a database coupled to said credit card issuer and said buyers computer;

updating selectively said history of previous signatures within said database; and

updating selectively said data within said database.

11. A method for fraud detection in an electronic transaction by a purchaser, including the steps of:

providing at a buyer'"'"'s computer an account number for a particular credit card account;

providing at said buyer'"'"'s computer an electronic signature capture device;

creating at said buyers computer, data representing a signature from said electronic signature capture device;

transmitting said account number and said data representing a signature from said seller to a credit card issuer over a computer network;

providing at said credit card issuer a multi-layer neural network combiner having N “

zero-one”

inputs and one “

zero-one”

output;

collecting at said credit card issuer within a database, training data for said neural network combiner, including the steps of recording a plurality of signature verification attempts and their-associated transactions, and keeping track of the transactions which are fraudulent transactions;

training at said credit card issuer said multi-layer neural network combiner on said N “

zero-one”

inputs and one “

zero-one”

output, wherein the N inputs are used to represent the probabilities output from each of the individual estimators which are to be combined using on-line backpropagation with targets of zero for those transactions which turned out to be fraudulent and one for those transactions which turned out to be valid;

evaluating at said credit card issuer, individual transactions by first generating the individual probabilities, and combining them by presenting the individual probabilities as inputs to the neural network and

transmitting an acceptance or non-acceptance of the transaction from said credit card issuer to said seller over a network; and

updating said database.