Method and apparatus for presenting feature importance in predictive modeling

• US 7,561,158 B2
• Filed: 01/11/2006
• Issued: 07/14/2009
• Est. Priority Date: 01/11/2006
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Tipu, Fateh Ali, Abe, Naoki, Pednault, Edwin Peter Dawson
• Primary Examiner(s): Amini; Javid A
• Application Number:
  US11/329,437
• Publication Number:
  US 20070159481A1
• Time in Patent Office: 1,280 Days
• Field of Search: 345/440, 706/12
• US Class Current: 345/440
• CPC Class Codes: G06T 11/206 Drawing of charts or graphs

Abstract

Feature importance information available in a predictive model with correlation information among the variables is presented to facilitate more flexible choices of actions by business managers. The displayed feature importance information combines feature importance information available in a predictive model with correlational information among the variables. The displayed feature importance information may be presented as a network structure among the variables as a graph, and regression coefficients of the variables indicated on the corresponding nodes in the graph. To generate the display, a regression engine is called on a set of training data that outputs importance measures for the explanatory variables for predicting the target variable. A graphical model structural learning module is called that outputs a graph on the explanatory variables of the above regression problem representing the correlational structure among them. The feature importance measure, output by the regression engine, is displayed for each node in the graph, as an attribute, such as color, size, texture, etc, of that node in the graph output by the graphical model structural learning module.

Claims

1. A method of displaying feature importance in predictive modeling comprising the steps of:

using a computer system having network connectivity to call a regression engine on a set of training data obtained from a storage unit connected to a computer network, said regression engine performing predictive modeling on said training data and outputting importance measures for explanatory variables for predicting a target variable;

calling a graphical model structural learning module that receives the importance measures output by the regression engine, computes correlational information among the explanatory variables, and outputs a graph on the explanatory variables and representing a feature correlation structure among said explanatory variables; and

displaying a feature importance measure, output by the regression engine, for each node in the graph, as an attribute of a node in the graph output by the graphical model structural learning module, to combine the predictive modeling of the regression engine with the feature correlation among the explanatory variables.

2. The method of displaying feature importance in predictive modeling recited in claim 1, wherein the feature importance measure, output by the regression engine, for each node in the graph, is displayed as an attribute selected from color, size, and texture.

3. The method of displaying feature importance in predictive modeling recited in claim 1, wherein the graphical model is obtained by calling a graphical model structure learning module using as input a correlation matrix between transforms of the input explanatory variables, in which the transform of each real valued variable is obtained by way of uni-variate linear regression tree with that input variable for the target variable, whereby the input explanatory variables are transformed so as to best predict the target variable.

4. The method of displaying feature importance in predictive modeling recited in claim 1, wherein the graphical model is obtained by calling a graphical model structure learning module using as input a correlation matrix between transforms of the input explanatory variables, in which the transform of each categorical variable is obtained by way of uni-variate regression tree with that input variable for the target variable, whereby the input explanatory variables are transformed so as to best predict the target variable.

5. The method of displaying feature importance in predictive modeling recited in claim 1, wherein the graphical model is obtained by calling an estimation method for dependency trees.

6. A system for displaying feature importance in predictive modeling comprising:

a regression engine operable on a set of training data that performs predictive modeling on the set of training data and outputs importance measures for explanatory variables for predicting a target variable;

a graphical model structural learning module that receives the importance measures output by the regression engine, computes correlational information among the explanatory variables, and outputs a graph on the explanatory variables and represents a feature correlation structure among said explanatory variables; and

a display for displaying a feature importance measure, output by the regression engine, for each node in the graph, as an attribute of a node in the graph output by the graphical model structural learning module, to combine the predictive modeling of the regression engine with the feature correlation among the explanatory variables.

7. The system for displaying feature importance in predictive modeling recited in claim 6, wherein the feature importance measure, output by the regression engine, for each node in the graph, is displayed by the display as an attribute selected from color, size, and texture.

8. The system for displaying feature importance in predictive modeling recited in claim 6, wherein the graphical model is obtained by calling the graphical model structure learning module using as input a correlation matrix between transforms of the input explanatory variables, in which the transform of each real valued variable is obtained by way of uni-variate linear regression tree with that input variable for the target variable, whereby the input explanatory variables are transformed so as to best predict the target variable.

9. The system for displaying feature importance in predictive modeling recited in claim 6, wherein the graphical model is obtained by calling the graphical model structure learning module using as input a correlation matrix between transforms of the input explanatory variables, in which the transform of each categorical variable is obtained by way of uni-variate regression tree with that input variable for the target variable, whereby the input explanatory variables are transformed so as to best predict the target variable.

10. The system for displaying feature importance in predictive modeling recited in claim 6, wherein the graphical model is obtained by calling an estimation method for dependency trees.

11. A storage unit having computer code implementing a method of displaying feature importance in predictive modeling, said method comprising the steps of:

calling a regression engine on a set of training data that performs predictive modeling on the set of training data and outputs importance measures for explanatory variables for predicting a target variable;

calling a graphical model structural learning module that receives the importance measures output by the regression engine, computes correlational information among the explanatory variables, and outputs a graph on the explanatory variables and representing a feature correlation structure among said explanatory variables; and

displaying a feature importance measure, output by the regression engine, for each node in the graph, as an attribute of a node in the graph output by the graphical model structural learning module, to combine the predictive modeling of the regression engine with the feature correlation among the explanatory variables.

12. The storage unit recited in claim 11, wherein the feature importance measure of the method of displaying feature importance in predictive modeling, output by the regression engine, for each node in the graph, is displayed as an attribute selected from color, size, and texture.

13. The storage unit recited in claim 11, wherein the graphical model of the method of displaying feature importance in predictive modeling is obtained by calling a graphical model structure learning module using as input a correlation matrix between transforms of the input explanatory variables, in which the transform of each real valued variable is obtained by way of uni-variate linear regression tree with that input variable for the target variable, whereby the input explanatory variables are transformed so as to best predict the target variable.

14. The storage unit recited in claim 11, wherein the graphical model of the method of displaying feature importance in predictive modeling is obtained by calling a graphical model structure learning module using as input a correlation matrix between transforms of the input explanatory variables, in which the transform of each categorical variable is obtained by way of uni-variate regression tree with that input variable for the target variable, whereby the input explanatory variables are transformed so as to best predict the target variable.

15. The storage unit recited in claim 11, wherein the graphical model of the method of displaying feature importance in predictive modeling is obtained by calling an estimation method for dependency trees.