Computer-implemented predictive model generation systems and methods
First Claim
Patent Images
1. A computer-implemented method for building a system of predictive models to predict credit card or debit card fraud, comprising:
- receiving, using one or more processors, training data including data items from a plurality of accounts;
training a first predictive model using the received training data;
using a partitioning criterion to determine how to partition the training data into partitions, wherein the partitioning criterion is based upon a fraud ranking violation metric;
partitioning the training data into partitions, wherein partitioning includes weighting data items contained in the training data differently based upon how well a data item learns during the training of the first predictive model; and
training additional predictive models using at least one of the partitions of training data in order to generate a second predictive model, wherein training of the additional predictive models optimizes the fraud ranking violation metric, and wherein the first and second predictive models are combined and used to predict credit card or debit card fraud.
1 Assignment
0 Petitions
Accused Products
Abstract
Systems and methods for performing fraud detection. As an example, a system and method can be configured to build a set of predictive models to predict credit card or debit card fraud. A first predictive model is trained using a set of training data. A partitioning criterion is used to determine how to partition the training data into partitions. Another predictive model is trained using at least one of the partitions of training data in order to generate a second predictive model. The predictive models are combined for use in predicting credit card or debit card fraud.
167 Citations
57 Claims
-
1. A computer-implemented method for building a system of predictive models to predict credit card or debit card fraud, comprising:
-
receiving, using one or more processors, training data including data items from a plurality of accounts; training a first predictive model using the received training data; using a partitioning criterion to determine how to partition the training data into partitions, wherein the partitioning criterion is based upon a fraud ranking violation metric; partitioning the training data into partitions, wherein partitioning includes weighting data items contained in the training data differently based upon how well a data item learns during the training of the first predictive model; and training additional predictive models using at least one of the partitions of training data in order to generate a second predictive model, wherein training of the additional predictive models optimizes the fraud ranking violation metric, and wherein the first and second predictive models are combined and used to predict credit card or debit card fraud. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14)
-
-
15. A computer-implemented method to generate reason information for a fraud score generated by a predictive model, said method comprising:
-
storing input data related to financial transactions; wherein the stored input data are associated with multiple different types of entities; generating a score based upon data regarding a new incremental transaction with respect to an entity; generating reason information associated with the score generated for the entity; wherein the reason information provides reasons for value of the score generated for the entity; wherein the generating of the reason information is based on fraud risk reason factors; wherein the fraud risk reason factors were generated by grouping similar fraud risk reason variables together; whereby the reason information is provided to a user for use in conducting fraud analysis. - View Dependent Claims (16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33)
-
-
34. A computer-implemented system to generate reason information for a fraud score generated by a predictive model, said method comprising:
-
financial transaction data store to store input data related to financial transactions; wherein the stored input data are associated with multiple different types of entities; score generation software instructions to generate a score based upon data regarding a new incremental transaction with respect to an entity; reason generation software instructions to generate reason information associated with the score generated for the entity; wherein the reason information provides reasons for value of the score generated for the entity; wherein the generating of the reason information is based on fraud risk reason factors; wherein the fraud risk reason factors were generated by grouping similar fraud risk reason variables together; whereby the reason information is provided to a user for use in conducting fraud analysis.
-
-
35. A computer-implemented method for building a system of predictive models to predict credit card or debit card fraud, comprising:
-
receiving, using one or more processors, training data including data items from a plurality of accounts; training a first predictive model using the received training data, wherein the first predictive model is trained at the account level; and
wherein the first predictive model considers change to compromised state at a transaction level;using a partitioning criterion to determine how to partition the training data into partitions, wherein the partitioning criterion is based upon a fraud ranking violation metric; and training additional predictive models using at least one of the partitions of training data in order to generate a second predictive model, wherein training of the additional predictive models optimizes the fraud ranking violation metric, and wherein the first and second predictive models are combined and used to predict credit card or debit card fraud. - View Dependent Claims (36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48)
-
-
49. A computer-implemented method for building a system of predictive models to predict credit card or debit card fraud, comprising:
-
receiving, using one or more processors, training data including data items from a plurality of accounts; training a first predictive model using the received training data; using a partitioning criterion to determine how to partition the training data into partitions, wherein the partitioning criterion is based upon a fraud ranking violation metric; partitioning the training data into partitions, wherein partitioning includes partitioning the training data into multiple data subsets; and training additional predictive models using at least one of the partitions of training data in order to generate a second predictive model, wherein training of the additional predictive models optimizes the fraud ranking violation metric, and wherein the first and second predictive models are combined and used to predict credit card or debit card fraud. - View Dependent Claims (50, 51)
-
-
52. A computer-implemented method for building a system of predictive models to predict credit card or debit card fraud, comprising:
-
receiving, using one or more processors, training data including data items from a plurality of accounts; training a first predictive model using the received training data; using a partitioning criterion to determine how to partition the training data into partitions, wherein the partitioning criterion is based upon a fraud ranking violation metric; partitioning the training data into partitions, wherein partitioning is based upon the fraud ranking violation metric; and training additional predictive models using at least one of the partitions of training data in order to generate a second predictive model, wherein training of the additional predictive models optimizes the fraud ranking violation metric, and wherein the first and second predictive models are combined and used to predict credit card or debit card fraud.
-
-
53. A computer-implemented method for building a system of predictive models to predict credit card or debit card fraud, comprising:
-
receiving, using one or more processors, training data including data items from a plurality of accounts; training a first predictive model using the received training data; using a partitioning criterion to determine how to partition the training data into partitions, wherein the partitioning criterion is based upon a fraud ranking violation metric; training additional predictive models using at least one of the partitions of training data in order to generate a second predictive model, wherein training of the additional predictive models optimizes the fraud ranking violation metric; and combining the first predictive model and second predictive models, wherein the combination is used to predict credit card or debit card fraud by performing a fraud scoring on demand process including; generating a first score on an account-level based upon data regarding a new incremental transaction with respect to an entity, subsequent to the generation of the first score, receiving a trigger that was generated without an incremental transaction having occurred, and in response to the generated trigger, generating a second score for the entity based upon stored past incremental transaction data and upon non-incremental transaction data, wherein the second score is indicative of whether fraud has occurred or not, and wherein the second score is used to determine whether a fraud handling action is performed upon the entity.
-
-
54. A system, comprising:
-
one or more processors; one or more computer-readable storage mediums containing instructions configured to cause the one or more processors to perform operations including; receiving training data including data items from a plurality of accounts; training a first predictive model using the received training data; using a partitioning criterion to determine how to partition the training data into partitions, wherein the partitioning criterion is based upon a fraud ranking violation metric; partitioning the training data into partitions, wherein partitioning includes weighting data items contained in the training data differently based upon how well a data item learns during the training of the first predictive model; and training additional predictive models using at least one of the partitions of training data in order to generate a second predictive model, wherein training of the additional predictive models optimizes the fraud ranking violation metric, and wherein the first and second predictive models are combined and used to predict credit card or debit card fraud.
-
-
55. A computer-program product, tangibly embodied in a machine-readable storage medium, including instructions operable to cause a data processing apparatus to:
-
receive training data including data items from a plurality of accounts; train a first predictive model using the received training data; use a partitioning criterion to determine how to partition the training data into partitions, wherein the partitioning criterion is based upon a fraud ranking violation metric; partition the training data into partitions, wherein partitioning includes weighting data items contained in the training data differently based upon how well a data item learns during the training of the first predictive model; and train additional predictive models using at least one of the partitions of training data in order to generate a second predictive model, wherein training of the additional predictive models optimizes the fraud ranking violation metric, and wherein the first and second predictive models are combined and used to predict credit card or debit card fraud.
-
-
56. A system, comprising:
-
one or more processors; one or more computer-readable storage mediums containing instructions configured to cause the one or more processors to perform operations including; receiving training data including data items from a plurality of accounts; training a first predictive model using the received training data, wherein the first predictive model is trained at the account level; and
wherein the first predictive model considers change to compromised state at a transaction level;using a partitioning criterion to determine how to partition the training data into partitions, wherein the partitioning criterion is based upon a fraud ranking violation metric; and training additional predictive models using at least one of the partitions of training data in order to generate a second predictive model, wherein training of the additional predictive models optimizes the fraud ranking violation metric, and wherein the first and second predictive models are combined and used to predict credit card or debit card fraud.
-
-
57. A computer-program product, tangibly embodied in a machine-readable storage medium, including instructions operable to cause a data processing apparatus to:
-
receive training data including data items from a plurality of accounts; train a first predictive model using the received training data, wherein the first predictive model is trained at the account level; and
wherein the first predictive model considers change to compromised state at a transaction level;use a partitioning criterion to determine how to partition the training data into partitions, wherein the partitioning criterion is based upon a fraud ranking violation metric; and train additional predictive models using at least one of the partitions of training data in order to generate a second predictive model, wherein training of the additional predictive models optimizes the fraud ranking violation metric, and wherein the first and second predictive models are combined and used to predict credit card or debit card fraud.
-
Specification