Method for generating predictive models in a computer system

• US 5,692,107 A
• Filed: 09/20/1996
• Issued: 11/25/1997
• Est. Priority Date: 03/15/1994
• Status: Expired due to Term

• Current Assignee: Lockheed Martin Corporation (Martin Marietta Corporation)
• Original Assignee: Lockheed Martin Space Operations Company (Martin Marietta Corporation)
• Inventors: Livezey, Brian K., Simoudis, Evangelos, Kerber, Randy G.
• Primary Examiner(s): Hafiz, Tariq R.
• Application Number:
  US08/718,226
• Time in Patent Office: 431 Days
• Field of Search: 395/10-13, 395/50-52, 395/54, 395/75-77, 395/600
• US Class Current: 706/12
• CPC Class Codes: G06N 5/025 Extracting rules from data

Abstract

Data mining system including a user interface 102, a plurality of data sources 114, at least one top-down data analysis module 104 and at least one bottom-up data analysis module 104'"'"' in cooperative communication with each other and with the user interface 102, and a server processor 106 in communication with the data sources 114 and with the data analysis modules 104, 104'"'"'. Data mining method involving the integration of top-down and bottom-up data mining techniques to extract 208 predictive models from a data source 114. A data source 114 is selected 200 and used to construct 202 a target data set 108. A data analysis module is selected 203 and module specific parameters are set 205. The selected data analysis module is applied 206 to the target data set based on the set parameters. Finally, predictive models are extracted 208 based on the target data set 108.

Claims

1. A data mining method for generating predictive models in a computer system, said computer system comprising:

a user interface;

at least one data source;

at least one top-down data analysis module and at least one bottom-up data analysis module in cooperative communication with each other and in communication with the user interface, where the top-down data analysis module considers data supporting and refuting a piece of expressed knowledge, validates or invalidates the knowledge, and gives reasons for the validity or invalidity of the knowledge, and the bottom-up data analysis module discovers knowledge in data; and

a server processor, in communication with each data source and with the data analysis modules;

the method comprising the steps of;

selecting data from at least one data source;

constructing a target data set from the data selected from the data source(s);

extracting a predictive model using at least one of the data analysis modules based on the target data set;

storing the predictive module for future use;

generating a knowledge base set, wherein said knowledge base set includes a set of rules, a validated query phrase, and said predictive model;

selecting the knowledge base set; and

validating a query phrase against the target data set and the knowledge base set;

wherein the step of extracting a predictive model comprises performing at least one process from the group of processes consisting of;

detecting a collection of rules and extracting the collection;

formulating a set of equations and extracting the set; and

training a neural network and extracting parameters describing the neural network.

2. A data mining method for generating predictive models in a computer system, said computer system comprising:

a user interface;

at least one data source;

at least one top-down data analysis module and at least one bottom-up data analysis module in cooperative communication with each other and in communication with the user interface, where the top-down data analysis module considers data supporting and refuting a piece of expressed knowledge, validates or invalidates the knowledge, and gives reasons for the validity or invalidity of the knowledge, and the bottom-up data analysis module discovers knowledge in data; and

a server processor, in communication with each data source and with the data analysis modules;

the method comprising the steps of;

selecting data from at least one data source;

constructing a target data set from the data selected from the data source(s);

extracting a predictive model using at least one of the data analysis modules based on the target data set;

storing the predictive module for future use;

generating a knowledge base set, wherein said knowledge base set includes a set of rules, a validated query phrase, and said predictive model;

selecting the knowledge base set; and

validating a query phrase against the target data set and the knowledge base set;

wherein the query phrase comprises a user-defined hypothesis, the method further comprising the steps of;

forming the hypothesis, using the data analysis module;

validating the hypothesis against the target data set; and

storing the validated hypothesis in the repository.

3. A data mining method for generating predictive models in a computer system, said computer system comprising:

a user interface;

at least one data source;

at least one top-down data analysis module and at least one bottom-up data analysis module in cooperative communication with each other and in communication with the user interface, where the top-down data analysis module considers data supporting and refuting a piece of expressed knowledge, validates or invalidates the knowledge, and gives reasons for the validity or invalidity of the knowledge, and the bottom-up data analysis module discovers knowledge in data; and

a server processor, in communication with each data source and with the data analysis modules;

the method comprising the steps of;

selecting data from at least one data source;

constructing a target data set from the data selected from the data source(s)i

extracting a predictive model using at least one of the data analysis modules based on the target data set; and

storing the predictive module for future use;

wherein at least one of the data sources comprises a relational database, the method further comprising the steps of;

extracting a schema of data, including tables and attributes, from the relational database;

defining the target data set including at least one table, having at least one of the attributes, from the schema;

defining a user query phrase using one of the data analysis modules;

validating a query phrase against the target data set;

storing the validated query phrase; and

selectively directing the validated query phrase to the server processor.

4. The method of claim 3, wherein the query phrase comprises a user-defined hypothesis, the method further comprising the steps of:

forming a hypothesis, using the top-down data analysis module;

validating the hypothesis against the target data set; and

storing the validated hypothesis in the repository.

5. The method of claim 3, wherein the data analysis modules include a visualization module, the method further comprising the step of generating a visual display of the validated query phrase at the user interface.

6. A data mining method for generating predictive models in a computer system, said computer system comprising:

a user interface;

at least one data source;

at least one top-down data analysis module and at least one bottom-up data analysis module in cooperative communication with each other and in communication with the user interface, where the top-down data analysis module considers data supporting and refuting a piece of expressed knowledge, validates or invalidates the knowledge, and gives reasons for the validity or invalidity of the knowledge, and the bottom-up data analysis module discovers knowledge in data; and

a server processor, in communication with each data source and with the data analysis modules;

the method comprising the steps of;

selecting data from at least one data source;

constructing a target data set from the data selected from the data source(s);

extracting a predictive model using at least one of the data analysis modules based on the target data set; and

storing the predictive module for future use;

wherein at least one of the data sources comprises a relational database, the method further comprising the steps of;

extracting a schema of data, including tables and attributes, from the relational database; and

defining the target data set including at least one table, having at least one of the attributes, from the schema;

wherein the step of defining the target data set includes the step of joining a plurality of the tables.

7. A data mining method for generating predictive models in a computer system, said computer system comprising:

a user interface;

at least one data source;

at least one top-down data analysis module and at least one bottom-up data analysis module in cooperative communication with each other and in communication with the user interface, where the top-down data analysis module considers data supporting and refuting. A piece of expressed knowledge, validates or invalidates the knowledge, and gives reasons for the validity or invalidity of the knowledge, and the bottom-up data analysis module discovers knowledge in data; and

a server processor, in communication with each data source and with the data analysis modules;

the method comprising the steps of;

selecting data from at least one data source;

constructing a target data set from the data selected from the data source(s);

extracting a predictive model using at least one of the data analysis modules based on the target data set; and

storing the predictive module for future use;

wherein at least one of the data sources comprises a relational database, the method further comprising the steps of;

extracting a schema of data, including tables and attributes, from the relational database; and

defining the target data set including at least one table, having at least one of the attributes, from the schema;

wherein the step of defining the target data set includes the step of constraining attributes of a selected table.

8. A data mining method for generating predictive models in a computer system, said computer system comprising:

a user interface;

at least one data source;

at least one top-down data analysis module and at least one bottom-up data analysis module in cooperative communication with each other and in communication with the user interface, where the top-down data analysis module considers data supporting and refuting a piece of expressed knowledge, validates or invalidates the knowledge, and gives reasons for the validity or invalidity of the knowledge, and the bottom-up data analysis module discovers knowledge in data; and

a server processor, in communication with each data source and with the data analysis modules;

the method comprising the steps of;

selecting data from at least one data source;

constructing a target data set from the data selected from the data source(s);

extracting a predictive model using at least one of the data analysis modules based on the target set; and

storing the predictive module for future use;

wherein the data analysis modules include an induction module, the method further comprising the steps of;

selecting the induction module using the user interface;

altering the target data set using user-specified parameters;

specifying a goal attribute; and

generating predictive modules in the form of rules.

9. The method of claim 8, further comprising the step of editing the set of rules in accordance with user-specified parameters using the user interface.

10. The method of claim 9, further comprising the step of storing the edited set of rules in the repository.

11. The method of claim 8, further comprising the step of testing the set of rules against the altered target data set.