Predicting data tampering using augmented machine learning models

Patent metadata

Field	Value
Publication number	US-2022103589-A1
Application number	US-202017037561-A
Country	US
Kind code	A1
Filing date	Sep 29, 2020
Priority date	Sep 29, 2020
Publication date	Mar 31, 2022
Grant date	—

Abstract

Certain aspects involve using a set of machine learning modeling models for predicting attempts to tamper with records using a fraudulent dispute. A tampering prediction system receives a request from a target entity to modify event data for a historical event, including information about the target entity and the event. The system generates a first score by applying a first set of machine learning models to the information from the request and information about the target entity obtained from a database. They system computes a second score by applying a second machine learning model to event data retrieved from the database. The second machine learning model has been trained using labeled training data and is augmented with a model that has been trained using unlabeled training data. The system generates an overall score for the request based on the first score and the second score.

First claim

1 . A system comprising a server computer comprising:
one or more processors; one or more non-transitory memories coupled to the one or more processors, the one or more memories storing: a database comprising event data for a plurality of historical events associated with a plurality of target entities; and a plurality of instructions executable by the one or more processors, the plurality of instructions comprising instructions that when executed by the one or more processors cause the one or more processors to perform processing comprising:
receiving a request from a target entity to modify event data for a historical event, the request comprising information about the target entity and information about the historical event;
generating a first security assessment score by, at least, applying a first set of machine learning models to the information about the target entity obtained from the request, the information about the historical event, and information about the target entity obtained from the database;
retrieving, from the database, event data associated with prior event data modification requests made by the target entity;
computing a second security assessment score by, at least, applying a second machine learning model to the retrieved event data, wherein the second machine learning model has been trained using labeled training data of the event data, and wherein the second machine learning model is augmented with an optimization model that has been trained using unlabeled training data of the event data;
generating an overall security assessment score for the request based on the first security assessment score and the second security assessment score; and
providing the overall security assessment score for the request to a client computer; and
the client computer, wherein the client computer is configured for preventing, based on the overall security assessment score, the target entity from accessing a resource.

2 . The system of claim 1 , wherein preventing the target entity from accessing the resource includes one or more of: preventing the target entity from accessing a secure database, preventing the target entity from accessing a secured portion of a website, or preventing the target entity from accessing a particular tool in an online environment.

3 . The system of claim 1 , the processing further comprising training the second machine learning model by:
obtaining the labeled training data; training the second machine learning model using the labeled training data; obtaining the unlabeled training data; training the optimization model using the unlabeled training data; and optimizing hyperparameters of the second machine learning model using the optimization model.

4 . The system of claim 1 , the processing further comprising:
identifying, from the database, additional historical event data associated with the outcome of prior requests made by the target entity; applying a third machine learning model to the additional historical event data to compute a third security assessment score, wherein the third machine learning model has been trained using the labeled training data, and wherein the third machine learning model is augmented with a second optimization model that has been trained using the unlabeled training data; and wherein the third machine learning model further bases the overall security assessment score on the third security assessment score.

5 . The system of claim 1 , wherein:
the request further includes information about a location of origin of the request; and applying the first set of machine learning models to generate the first security assessment score further comprises:
applying a fourth machine learning model to the location information to generate first risk signals; and
applying a fifth machine learning model to the information about the target entity obtained from the database and the information about the target entity obtained from the request to generate second risk signals;
wherein the first security assessment score is generated by applying a sixth machine learning model to the first risk signals and the second risk signals.

6 . The system of claim 5 , wherein:
the information about the historical event includes a narrative description of the historical event; and applying the first set of machine learning models to generate the first security assessment score further comprises: executing optical character recognition on the narrative description of the historical event to extract information from the narrative description of the historical event; and
performing natural language processing on the extracted information to generate third risk signals,
wherein the first security assessment score is further based on the third risk signals.

7 . The system of claim 5 , wherein:
the request further includes a supplemental document; and applying the first set of machine learning models to generate the first security assessment score further comprises: executing optical character recognition on the supplemental document to extract information from the supplemental document; and
based on the information extracted from the supplemental document, generating fourth risk signals,
wherein the first security assessment score is further based on the fourth risk signals.

8 . The system of claim 5 , wherein:
the sixth machine learning model has been trained using labeled risk signal data, and wherein the sixth machine learning model is augmented with an optimization model that has been trained using unlabeled risk signal data.

9 . A computer-implemented method comprising:
receiving, by a server computer, a request from a target entity to modify event data for a historical event, of a plurality of historical events associated with a plurality of target entities stored to a database, the request comprising information about the target entity and information about the historical event; generating, by the server computer, a first security assessment score by, at least, applying a first set of machine learning models to the information about the target entity obtained from the request, the information about the historical event, and information about the target entity obtained from the database; retrieving, by the server computer from the database, event data associated with prior requests made by the target entity; computing, by the server computer, a second security assessment score by at least applying a second machine learning model to the retrieved event data associated with the prior requests made by the target entity, wherein the second machine learning model has been trained using labeled training data of the event data, and wherein the second machine learning model is augmented with an optimization model that has been trained using unlabeled training data of the event data; generating, by the server computer, an overall security assessment score for the request based on the first security assessment score and the second security assessment score; and providing, by the server computer, the overall security assessment score for the request to a client computer, wherein the overall security assessment score is usable by the client computer for preventing the target entity from accessing a resource.

10 . The method of claim 9 , wherein preventing the target entity from accessing the resource includes one or more of: preventing the target entity from accessing a secure database, preventing the target entity from accessing a secured portion of a website, or preventing the target entity from accessing a particular tool in an online environment.

11 . The method of claim 9 , further comprising
determinin

Assignees

• Equifax Inc

Inventors

• Shen Cuizhen
• Munguia Philip
• Agrawal Prateek
• Chen Ledao
• Tirunellayi Sriram

Classifications

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06Q40/03

  Credit; Loans; Processing thereof · CPC title

• G06N3/09

  Supervised learning · CPC title