Method and System of Predictive Document Verification and Machine Learning Therefor
Provided are methodology and system countering fraudulent document and/or image use when authentication of a transaction based on a given document or image use is required. Additionally provided is a manner of machine learning adapting the methodology for implementation thereof.
1 . A method of verifying an identity of an individual for authenticating a transaction, the method comprising:
receiving, as offered proof of identity of the individual for the transaction, a selfie of the individual comprising one or more elements and/or a document expression for a presented document of the individual comprising one or more elements, the document expression comprising an image of the presented document which comprises at least a headshot of the individual and identity information of the individual comprising personally identifiable information (PII) comprising at least a name and a date of birth (DOB);
determining, by an identity verifier (IV), an evaluation of fraudulent usage for (a) the selfie of the individual and/or (b) the document expression by at least cross-checking the document expression against a known standard for the presented document to evaluate compliance with the standard; and
identifying, based on the evaluation, an authentication result for the transaction comprising a probability that the transaction is fraudulent, by:
creating a first training set comprising one or more of (i) training selfies of individuals and (ii) training document expressions, in which elements comprising a training selfie and a training document expression are each respectively initially paired with a predetermined fraud weighting,
wherein each predetermined fraud weighting indicates a respective probability of fraud in connection with use of the selfie of the individual or the document expression;
converting the evaluation into input for a machine learning model comprising an identity verification predictor (IVP) trained on the first training set;
applying the input to the IVP and, obtaining, as output from the IVP, the authentication result for the transaction,
wherein the authentication result is, in response to comparison by the IVP of the first training set to the one or more elements of the selfie of the individual and/or the one or more elements of the document expression, based on rankings corresponding to fraud risk weightings assigned by the IVP to the compared one or more elements of the selfie of the individual and/or the document expression, wherein the rankings comprise a hierarchy according to the corresponding fraud risk weightings respectively assigned to the one or more elements of the selfie and/or the one or more elements of the document expression;
retaining the IVP by tuning the assigned fraud risk weightings, in response to the IVP identifying a ranking insufficiency among (a) at least a pair of the elements of the selfie (b) at least a pair of the elements of the document expression, or (c) at least a pair of elements derived from both the selfie and the document expression, and
verifying, based on the retaining of the IVP, the authentication result,
wherein, in response to the retraining and the verifying, the hierarchy comprises an adjusted ordering of the elements of the selfie and/or the elements of the document comprises an adjusted ordering of the elements of the selfie and/or the elements of the document expression based on a normalized distribution for the rankings.
2 . The method of claim 1 ,
wherein the evaluation of the selfie comprises one or more of (a) determining liveness and/or spoofing thereof, (b) determining an age thereof, (c) determining whether the depiction for the selfie image appears in one or more data stores comprising images for individuals associated with fraudulent activity, or (d) any combination thereof.
3 . The method of claim 1 ,
wherein the known standard for the presented document comprises one or more of (a) embeddings, (b) placement, sizing, and/or spacing for the PII, (c) presentation of encoded data comprising a machine-readable version of the PII, or (d) any combination thereof.
4 . The method of claim 1 ,
wherein the evaluation of the presented document further comprises cross-checking the PII with one or more data stores to verify the PII and/or determine a frequency of presentation as to a misrepresentation of the PII, and/or a comparison between a mathematical representation of the headshot with mathematical representations of headshots correlated to the PII as included in one or more data stores.
5 . The method of claim 1 ,
wherein the evaluation further comprises a comparison between (a) the selfie and the headshot to determine matching therebetween and/or (b) a comparison of an estimated age of the selfie and the age of the individual as determined by the DOB.
6 . The method of claim 1 ,
wherein the evaluation further comprises comparing one or more features of the selfie and/or the headshot to respective model scores for the one or more features.
7 . The method of claim 1 ,
wherein the evaluation further comprises determining a level of one or more photographic distortions for the selfie and/or the headshot.
8 . The method of claim 1 ,
wherein the evaluation further comprises determining, based on a comparison with headshots and corresponding PII as included in one or more data stores, (a) the presence or absence of physical traits of the individual as presented on the selfie and/or the headshot and/or (b) incorrect PII usage by the individual in connection with the selfie and/or the headshot.
9 . The method of claim 1 ,
wherein the evaluation further comprises obtaining identifying information of a device used to capture the selfie, and determining, based on data of one or more data stores, whether the identifying information has been previously used in connection with fraudulent use of the selfie.
10 . A computing system for verifying an identity of an individual to authenticate a transaction, the computing system comprising:
one or more processors;
one or more memories storing instructions that, when executed by the one or more
processors, cause the computing system to perform a process comprising:
receiving, as offered proof of identity of the individual for the transaction, a selfie of the individual comprising one or more elements and/or a document expression for a presented document of the individual comprising one or more elements, the document expression comprising an image of the presented document which comprises at least a headshot of the individual and identity information of the individual comprising personally identifiable information (PII) comprising at least a name and a date of birth (DOB);
determining, by an identity verifier (IV), an evaluation of fraudulent usage for (a) the selfie of the individual and/or (b) the document expression by at least cross-checking the document expression against a known standard for the presented document to evaluate compliance with the standard; and
identifying, based on the evaluation, an authentication result for the transaction comprising a probability that the transaction is fraudulent, by:
creating a first training set comprising one or more of (i) training selfies of individuals and (ii) training document expressions, in which elements comprising a training selfie and a training document expression are each respectively initially paired with a predetermined fraud weighting,
wherein each predetermined fraud weighting indicates a respective probability of fraud in connection with use of the selfie of the individual or the document expression;
converting the evaluation into input for a machine learning model comprising an identity verification predictor (IVP) trained on the first training set;
applying the input to the IVP and, obtaining, as output from the IVP, the authentication result for the transaction,
wherein the authentication result is, in response to comparison by the IVP of the first training set to the one or more elements of the selfie of the individual and/or the one or more elements of the document expression, based on rankings corresponding to fraud risk weightings assigned by the IVP to the compared one or more elements of the selfie of the individual and/or the document expression, wherein the rankings comprise a hierarchy according to the corresponding fraud risk weightings respectively assigned to the one or more elements of the selfie and/or the one or more elements of the document expression;
retraining the IVP by tuning the assigned fraud risk weightings, in response to the IVP identifying a ranking insufficiency among (a) at least a pair of the elements of the selfie, (b) at least a pair of the elements of the document expression, or (c) at least a pair of elements derived from both the selfie and the document expression: and
verifying, based on the retraining of the IVP, the authentication result,
wherein, in response to the retraining and the verifying, the hierachy comprises an adjusted ordering of the elements of the selfie and/or the elements of the document expression based on a normalized distribution for the rankings.
11 . The computing system of claim 10 ,
wherein the evaluation of the selfie comprises one or more of (a) determining liveness and/or spoofing thereof, (b) determining an age thereof, (c) determining whether the depiction for the selfie image appears in one or more data stores comprising images for individuals associated with fraudulent activity, or (d) any combination thereof.
12 . The computing system of claim 10 ,
wherein the known standard for the presented document comprises one or more of (a) embeddings, (b) placement, sizing, and/or spacing for the PII, (c) presentation of encoded data comprising a machine-readable version of the PII, or (d) any combination thereof.
13 . The computing system of claim 10 ,
wherein the evaluation of the presented document further comprises cross-checking the PII with one or more data stores to verify the PII and/or determine a frequency of presentation as to a misrepresentation of the PII, and/or a comparison between a mathematical representation of the headshot with mathematical representations of headshots correlated to the PII as included in one or more data stores.
14 . The computing system of claim 10 ,
wherein the evaluation further comprises a comparison between (a) the selfie and the headshot to determine matching therebetween and/or (b) a comparison of an estimated age of the selfie and the age of the individual as determined by the DOB.
15 . The computing system of claim 10 ,
wherein the evaluation further comprises comparing one or more features of the selfie and/or the headshot to respective model scores for the one or more features.
16 . The computing system of claim 10 ,
wherein the evaluation further comprises determining a level of one or more photographic distortions for the selfie and/or the headshot.
17 . The computing system of claim 10 ,
wherein the evaluation further comprises determining, based on a comparison with headshots and corresponding PII as included in one or more data stores, (a) the presence or absence of physical traits of the individual as presented on the selfie and/or the headshot and/or (b) incorrect PII usage by the individual in connection with the selfie and/or the headshot.
18 . The computing system of claim 10 ,
wherein the evaluation further comprises obtaining identifying information of a device used to capture the selfie, and determining, based on data of one or more data stores, whether the identifying information has been previously used in connection with fraudulent use of the selfie.