# US 2018/0025104 A1 · Automatic Modeling Farmer > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2018/0025104 A1 - Document ID: US20180025104A1 - Status: Published application - Kind: A1 - Application number: 15/722,143 - Filed: 2017-10-02 - Published: 2018-01-25 - Earliest claimed: 2013-08-26 - Assignees: Fair Isaac Corporation - Applicants: Fair Isaac Corporation - Inventors: Lu Gao; Brendan Alexander Lacounte; Jane Sherman; Radha Chandra; Erik Franco; Ethan J. Dornhelm - Web page: https://patentgrep.com/patent/US20180025104A1 ## CPC - G06F 16/256: {in federated or virtual databases} - G06F 16/212: {with details for data modelling support} - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) ## Abstract Data can be accessed from a plurality of disparate data sources from at least one database. A plurality of test models can be automatically built by a model building engine. Each test model can have predetermined predictive variables. A final set of predictive variables can be determined by a variable selector from the predetermined predictive variables in the plurality of test models by comparing the predictive power of the predictive variables across the plurality of test models. A master dataset can be generated from the disparate data sources. A master model can be built from the master dataset. The master model can combine the final set of predictive variables from the plurality of disparate data sources. The master model can characterize a quantitative estimate of the probability that an entity will display a defined behavior. Related apparatus, systems, techniques, and articles are also described. ## Claims ### Claim 1 A method for implementation by one or more data processors forming part of at least one computing system, the method comprising: accessing, using at least one data processor and from at least one database, data from a plurality of disparate data sources; automatically building, by a model building engine using at least one data processor and the data obtained from the accessed data sources, a plurality of test models, each test model having predetermined predictive variables and each test model built from one or more of the plurality of disparate data sources; determining, by a variable selector and using at least one data processor, a final set of predictive variables from the predetermined predictive variables in the plurality of test models by comparing the predictive power of the predictive variables across the plurality of test models, the final set of predictive variables being the most predictive of the predetermined predictive variables; generating, using at least one data processor, a master dataset comprising data selected from the disparate data sources and corresponding to the determined final set of predictive variables; and building, using at least one data processor and from the master dataset, a master model that combines the final set of predictive variables from the plurality of disparate data sources, the master model characterizing a quantitative estimate of the probability that an entity will display a defined behavior. ### Claim 2 (depends on 1) The method of claim 1, further comprising iteratively building test models and determining an interim set of predictive variables until the test models or the interim set of predictive variables satisfy a predetermined criterion. ### Claim 3 (depends on 1) The method of claim 1, further comprising: breaking at least one of the disparate data sources into multiple child datasets so each child dataset contains only a subset of the predictive variables and the child datasets share the same performance variable, sampling scheme, and segmentation scheme; and building test models for each child dataset. ### Claim 4 (depends on 1) The method of claim 1, further comprising segmenting the disparate data source populations into more homogenous sub-populations. ### Claim 5 (depends on 1) The method of claim 1, further comprising merging data across multiple disparate data sources and in one or more varying combinations of the disparate data sources. ### Claim 6 (depends on 1) The method of claim 1, further comprising automatic down sampling of one or more disparate data sources when a number of records in the disparate data source exceeds a predetermined value. ### Claim 7 (depends on 1) The method of claim 1 wherein the disparate data sources include one or more of a credit and debit bureau database; financial institution database comprising records of financial transactions; personal data database regarding demographic, lifestyle and public records; social network database, and online usage database. ### Claim 8 (depends on 1) The method of claim 1, wherein the predictive variables are predetermined using a consumer behavior framework and include whether an individual has had one or more of: a marriage, family expansion, new job, additional income, a number of recent hard credit inquiries, and risk score. ### Claim 9 (depends on 1) The method of claim 1, further comprising reporting a compilation of user selected model performance metrics across all test models built in a user-defined format. ### Claim 10 A system comprising at least one data processor, and memory storing instructions which, when executed by the at least one data processor, causes the at least one data processor to perform operations comprising: accessing, using at least one data processor and from at least one database, data from a plurality of disparate data sources; automatically building, by a model building engine using at least one data processor and the data obtained from the accessed data sources, a plurality of test models, each test model having predetermined predictive variables and each test model built from one or more of the plurality of disparate data sources; determining, by a variable selector and using at least one data processor, a final set of predictive variables from the predetermined predictive variables in the plurality of test models by comparing the predictive power of the predictive variables across the plurality of test models, the final set of predictive variables being the most predictive of the predetermined predictive variables; generating, using at least one data processor, a master dataset comprising data selected from the disparate data sources and corresponding to the determined final set of predictive variables; and building, using at least one data processor and from the master dataset, a master model that combines the final set of predictive variables from the plurality of disparate data sources, the master model characterizing a quantitative estimate of the probability that an entity will display a defined behavior. ### Claim 11 (depends on 10) The system of claim 10, the operations further comprising iteratively building test models and determining an interim set of predictive variables until the test models or the interim set of predictive variables satisfy a predetermined criterion. ### Claim 12 (depends on 10) The system of claim 10, the operations further comprising: breaking at least one of the disparate data sources into multiple child datasets so each child dataset contains only a subset of the predictive variables and the child datasets share the same performance variable, sampling scheme, and segmentation scheme; and building test models for each child dataset. ### Claim 13 (depends on 10) The system of claim 10, the operations further comprising segmenting the disparate data source populations into more homogenous sub-populations. ### Claim 14 (depends on 10) The system of claim 10, the operations further comprising merging data across multiple disparate data sources and in one or more varying combinations of the disparate data sources. ### Claim 15 A non-transitory computer program product storing instructions, which when executed by at least one data processor of at least one computing system, implement a method, the method comprising: accessing, using at least one data processor and from at least one database, data from a plurality of disparate data sources; automatically building, by a model building engine using at least one data processor and the data obtained from the accessed data sources, a plurality of test models, each test model having predetermined predictive variables and each test model built from one or more of the plurality of disparate data sources; determining, by a variable selector and using at least one data processor, a final set of predictive variables from the predetermined predictive variables in the plurality of test models by comparing the predictive power of the predictive variables across the plurality of test models, the final set of predictive variables being the most predictive of the predetermined predictive variables; generating, using at least one data processor, a master dataset comprising data selected from the disparate data sources and corresponding to the determined final set of predictive variables; and building, using at least one data processor and from the master dataset, a master model that combines the final set of predictive variables from the plurality of disparate data sources, the master model characterizing a quantitative estimate of the probability that an entity will display a defined behavior. ### Claim 16 (depends on 15) The computer program product of claim 15, the method further comprising iteratively building test models and determining an interim set of predictive variables until the test models or the interim set of predictive variables satisfy a predetermined criterion. ### Claim 17 (depends on 15) The computer program product of claim 15, the method further comprising: breaking at least one of the disparate data sources into multiple child datasets so each child dataset contains only a subset of the predictive variables and the child datasets share the same performance variable, sampling scheme, and segmentation scheme; and building test models for each child dataset. ### Claim 18 (depends on 15) The computer program product of claim 15, the method further comprising segmenting the disparate data source populations into more homogenous sub-populations. ### Claim 19 (depends on 15) The computer program product of claim 15, the method further comprising merging data across multiple disparate data sources and in one or more varying combinations of the disparate data sources. ### Claim 20 (depends on 15) The computer program product of claim 15, the method further comprising automatic down sampling of one or more disparate data sources when a number of records in the disparate data source exceeds a predetermined value.