Method and apparatus for real-time fraud machine learning model execution module
Uloženo v:
| Název: | Method and apparatus for real-time fraud machine learning model execution module |
|---|---|
| Patent Number: | 11663,602 |
| Datum vydání: | May 30, 2023 |
| Appl. No: | 16/413085 |
| Application Filed: | May 15, 2019 |
| Abstrakt: | Various methods, apparatuses, and media for implementing a fraud machine learning model execution module are provided. A processor generates a plurality of machine learning models. The processor generates historical aggregate data based on prior transaction activities of a customer from a plurality of databases for transactions. The processor also tracks activities of the customer during a new transaction authorization process and generates a transaction data; integrates the transaction data with the historical aggregate data; executes each of said machine learning models using the integrated transaction data and the historical aggregate data to generate a fraud score and stores the fraud score into the memory; and determines whether the new transaction is fraudulent based on the generated fraud score. |
| Inventors: | JPMorgan Chase Bank, N.A. (New York, NY, US) |
| Assignees: | JPMORGAN CHASE BANK, N.A. (New York, NY, US) |
| Claim: | 1. A method for implementing a real-time fraud machine learning module (RTFMLM) to detect fraudulent transactions by utilizing one or more processors and one or more memories, the method comprising: implementing, by one or more processors, the RTFMLM in servers that are distributed across different communication networks; and configuring, by the one or more processors, the RTFMLM to: generate, via an open source framework, a plurality of machine learning models, each of said plurality of machine learning models is configured to run simultaneously in parallel and independent of each other; generate historical aggregate data based on prior transaction activities of a customer from a plurality of databases for transactions; in real-time, track activities of the customer during a new transaction authorization process of an operational system and utilize data of the new transaction authorization process to generate a transaction data, wherein real-time is any amount of time that is less than 130 milliseconds, and wherein the operational system includes at least one operational system from among a mainframe authorization decision engine (MF ADE), an automated credit application processing system (ACAPS), BMWNPC, and Taser; integrate the transaction data with the historical aggregate data; automatically update, via the open source framework, each of said plurality of machine learning models by incorporating model changes, due to the integration of the transaction data with the historical aggregate data, without recoding the model changes; generate, via the open source framework, new machine learning models based on the model changes; store at least one of each of the new machine learning models onto one or more of the servers that are distributed across the different communication networks; execute at least one of the new machine learning models in a simulation mode; after executing the at least one of the new machine learning models in the simulation mode more than twice, determine whether the at least one of the new machine learning models is running as expected; in response to determining that the at least one of the new machine learning models is running as expected, integrate the at least one of the new machine learning models into the RTFMLM and utilize the at least one of the new machine learning models to retrain at least one of said plurality of machine learning models; execute said plurality of machine learning models using the integrated transaction data and the historical aggregate data to generate a fraud score; and determine whether a new transaction is fraudulent in real-time and based on the generated fraud score, wherein: the new transaction is authorized when the fraud score is a value that is at or above a predetermined threshold, and the new transaction is denied when the fraud score is a value that is below the predetermined threshold. |
| Claim: | 2. The method according to claim 1 , wherein the generation of said plurality of machine learning models comprises: using predictive model markup language (PMML) as an open source framework to model each of said plurality of machine learning models. |
| Claim: | 3. The method according to claim 1 , wherein the generation of said plurality of machine learning models comprises: using any one of the following as an open source framework to model each of said plurality of machine learning models: Java Spring Boot, Cassandra, LogStash, Kibana, and Kafka. |
| Claim: | 4. The method according to claim 1 , further comprising configuring, by the one or more processors, the RTFMLM to: update each of said plurality of machine learning models by automatically incorporating model changes by using a predictive model markup language (PMML) as an open source framework; generate a new model based on the updating; and execute the new model in a simulation mode prior to executing each of said plurality of machine learning models. |
| Claim: | 5. The method according to claim 1 , further comprising configuring, by the one or more processors, the RTFMLM to: update each of said plurality of machine learning models by automatically incorporating model changes by using a predictive model markup language (PMML) as an open source framework; generate a new model based on the updating; store the new model onto a memory; and utilize the new model to retrain each of said plurality of machine learning models. |
| Claim: | 6. A system for implementing a real-time fraud machine learning module (RTFMLM) to detect fraudulent transactions, the system comprising: one or more processors; and a memory, wherein the one or more processors is configured to implement the RTFMLM in server devices that are distributed across different communication networks, and the RTFMLM is configured to: generate, via an open source framework, a plurality or machine learning models, each of said plurality of machine learning models is configured to run simultaneously in parallel and independent of each other; generate historical aggregate data based on prior transaction activities of a customer from a plurality of databases for transactions; in real-time, track activities of the customer during a new transaction authorization process of an operational system and utilize data of the new transaction authorization process to generate a transaction data, wherein real-time is any amount of time that is less than 130 milliseconds, and wherein the operational system includes at least one operational system from among a mainframe authorization decision engine (MF ADE), an automated credit application processing system (ACAPS), BMWNPC, and Taser; integrate the transaction data with the historical aggregate data; automatically update, via the open source framework, each of said plurality of machine learning models by incorporating model changes, due to the integration of the transaction data with the historical aggregate data, without recoding the model changes; generate, via the open source framework, new machine learning models based on the model changes; store at least one of each of the new machine learning models onto one or more of the servers that are distributed across the different communication networks; execute at least one of the new machine learning models in a simulation mode; after executing the at least one of the new machine learning models in the simulation mode more than twice, determine whether the at least one of the new machine learning models is running as expected; in response to determining that the at least one of the new machine learning models is running as expected, integrate the at least one of the new machine learning models into the RTFMLM and utilize the at least one of the new machine learning models to retrain at least one of said plurality of machine learning models; execute said plurality of machine learning models using the integrated transaction data and the historical aggregate data to generate a fraud score and store the fraud score into the memory; and determine whether a new transaction is fraudulent in real-time and based on the generated fraud score, wherein: the new transaction is authorized when the fraud score is a value that is at or above a predetermined threshold, and the new transaction is denied when the fraud score is a value that is below the predetermined threshold. |
| Claim: | 7. The system according to claim 6 , wherein the RTFMLM is further configured to: model each of said plurality of machine learning models by using predictive model markup language (PMML) as an open source framework. |
| Claim: | 8. The system according to claim 6 , wherein the RTFMLM is further configured to: model each of said plurality of machine learning models by using any one of the following as an open source framework: Java Spring Boot, Cassandra, LogStash, Kibana, and Kafka. |
| Claim: | 9. The system according to claim 6 , wherein the RTFMLM is further configured to: update each of said plurality of machine learning models by automatically incorporating model changes by using a predictive model markup language (PMML) as an open source framework; generate a new model based on the updating; and execute the new model in a simulation mode prior to executing each of said plurality of machine learning models. |
| Claim: | 10. The system according to claim 6 , wherein the RTFMLM is further configured to: update each of said plurality of machine learning models by automatically incorporating model changes by using a predictive model markup language (PMML) as an open source framework; generate a new model based on the updating; store the new model onto the memory; and utilize the new model to retrain each of said plurality of machine learning models. |
| Claim: | 11. A non-transitory computer readable medium configured to store instructions for implementing a real-time fraud machine learning module (RTFMLM) to detect fraudulent transactions, wherein when executed, the instructions cause one or more processors to perform the following: implementing the RTFMLM in servers that are distributed across different communication networks; and configuring the RTFMLM to perform the following: generating, via an open source framework, a plurality of machine learning models, each of said plurality of machine learning models is configured to run simultaneously in parallel and independent of each other; generating historical aggregate data based on prior transaction activities of a customer from a plurality of databases for transactions; in real-time, tracking activities of the customer during a new transaction authorization process of an operational system and utilizing data of the new transaction authorization process to generate a transaction data, wherein real-time is any amount of time that is less than 130 milliseconds, and wherein the operational system includes at least one operational system from among a mainframe authorization decision engine (MF ADE), an automated credit application processing system (ACAPS), BMWNPC, and Taser; integrating the transaction data with the historical aggregate data; automatically updating, via the open source framework, each of said plurality of machine learning models by incorporating model changes, due to the integration of the transaction data with the historical aggregate data, without recoding the model changes; generating, via the open source framework, new machine learning models based on the model changes; storing at least one of each of the new machine learning models onto one or more of the servers that are distributed across the different communication networks; executing at least one of the new machine learning models in a simulation mode; after executing the at least one of the new machine learning models in the simulation mode more than twice, determining whether each of the new machine learning models is running as expected; in response to determining that the at least one of the new machine learning models is running as expected, integrating the at least one of the new machine learning models into the RTFMLM and utilizing the at least one of the new machine learning models to retrain at least one of said plurality of machine learning models; executing said plurality of machine learning models using the integrated transaction data and the historical aggregate data to generate a fraud score; and determining whether a new transaction is fraudulent in real-time and based on the generated fraud score, wherein: the new transaction is authorized when the fraud score is a value that is at or above a predetermined threshold, and the new transaction is denied when the fraud score is a value that is below the predetermined threshold. |
| Claim: | 12. The non-transitory computer readable medium of claim 11 , wherein the instructions, when executed, further cause the one or more processors to configure the RTFMLM to perform the following: using predictive model markup language (PMML) as an open source framework to model each of said plurality of machine learning models. |
| Patent References Cited: | 10223691 March 2019 Katzin et al. 20110055078 March 2011 Nandy 20110295722 December 2011 Reisman 20150170147 June 2015 Geckle et al. 20160086185 March 2016 Adjaoute 20170364918 December 2017 Malhotra 20190095924 March 2019 Adjaoute 20200252802 August 2020 Pachauri |
| Other References: | Official communication (Search Report and Opinion) in W.I.P.O Patent Application No. PCT/US2020/32951, dated Jul. 29, 2020. cited by applicant |
| Assistant Examiner: | Werner, Marshall L |
| Primary Examiner: | Afshar, Kamran |
| Attorney, Agent or Firm: | Greenblum & Bernstein, P.L.C. |
| Přístupové číslo: | edspgr.11663602 |
| Databáze: | USPTO Patent Grants |
| Abstrakt: | Various methods, apparatuses, and media for implementing a fraud machine learning model execution module are provided. A processor generates a plurality of machine learning models. The processor generates historical aggregate data based on prior transaction activities of a customer from a plurality of databases for transactions. The processor also tracks activities of the customer during a new transaction authorization process and generates a transaction data; integrates the transaction data with the historical aggregate data; executes each of said machine learning models using the integrated transaction data and the historical aggregate data to generate a fraud score and stores the fraud score into the memory; and determines whether the new transaction is fraudulent based on the generated fraud score. |
|---|