Incremental effect modeling by area index maximization

• US 8,818,920 B2
• Filed: 03/09/2012
• Issued: 08/26/2014
• Est. Priority Date: 03/09/2012
• Status: Active Grant

• Current Assignee: Bank of America Corp.
• Original Assignee: Bank of America Corp.
• Inventors: Liu, Xiaohu, Li, Jing
• Primary Examiner(s): Gaffin, Jeffrey A.
• Assistant Examiner(s): Kim, David H
• Application Number:
  US13/416,149
• Publication Number:
  US 20130238539A1
• Time in Patent Office: 900 Days
• Field of Search: 706/46
• US Class Current: 706/12
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06Q 10/06 Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
• CPC Class Codes: G06Q 10/067 Enterprise or organisation modelling
• CPC Class Codes: G06Q 30/00 Commerce

Abstract

How much is the net benefit of treating an individual: cost-effective, very small, or even negative? To address this question, a methodology for developing an incremental effect model based on randomized test data is provided. The concept of an incremental effect area index is introduced for measuring the model'"'"'s quality. A new variable screening technique is proposed to identify variables that are decision relevant and preferably time invariant. A piecewise linear function is created for each continuous variable to approximate the relationship between the incremental effect and the variable, based on a binning technique. Finally, a score is created as the weighted sum of a set of functions, with each function being the empirical prediction of the incremental effect based on a variable, wherein weights are chosen to maximize the incremental effect area index. The methodology creates an improved incremental effect model, leading to more cost-effective strategies in business practice.

Claims

1. An incremental effect modeling apparatus comprising:

a computing device processor;

a storage device; and

an incremental effect modeling application stored in said storage device and configured for operation on said computing device processor, said application configured to;

determine a set of model variables from a plurality of predictor variables that are decision relevant and optionally time invariant, wherein the predictor variables are associated with a response variable or a performance variable and a decision variable for a population, wherein the decision variable defines treated observations and control observations in the population;

determine a function describing a relationship between an incremental effect and the model variable, wherein the incremental effect is defined as the difference between performance of an individual under treatment and the performance of the individual under no treatment;

rank order the treated and control observations in the population based on a score from low to high, wherein the score is a weighted sum of a set of functions created from the model variables;

determine a cumulative incremental effect at increasing percentages of ranked observations with the lowest score values, based on the combined test and control data;

determine an incremental effect area index based on the cumulative incremental effect as a function of the increasing percentages of ranked observations with the lowest score values; and

iterate through the functions of the model variables for weight searching until the incremental effect area index is maximized to determine the incremental effect model.

2. The incremental effect modeling apparatus of claim 1, the apparatus further comprising a communications module configured to communicate with a database.

3. The incremental effect modeling apparatus of claim 2, wherein the database comprises the plurality of predictor variables, response variables, and decision variables.

4. The incremental effect modeling apparatus of claim 1, the apparatus further comprising a network adapter configured to communicate with at least one of a user, a financial institution, and a marketing database.

5. The incremental effect modeling apparatus of claim 1, the apparatus further comprising a marketing module, wherein the marketing module is configured to treat individuals based on the incremental effect model determined by the incremental effect modeling application.

6. The incremental effect modeling apparatus of claim 5, wherein the marketing module automatically contacts the individuals.

7. The incremental effect modeling apparatus of claim 1, wherein the application is further configured to rank order the model variables based on a statistical test of the significance of the two-term interaction between the model variable and the decision variable according to a linear model, wherein the model variable and the decision variable are treated as class variables, and the response variable is regressed on the model variable, the decision variable, and the interaction between the model variable and the decision variable.

8. The incremental effect modeling apparatus of claim 1, wherein the application further comprises an area index maximization module configured to analyze multiple model scores comprising weighted sum of nonlinear functions and select the score wherein the incremental effect area index is maximized.

9. The incremental effect modeling apparatus of claim 1, wherein predictor variables are excluded if a three-term interaction of the decision variable, the predictor variable, and a time variable has a p-value lower than a predetermined p-value, wherein the three-term interaction is determined based on a linear model which treats the decision variable, the predictor variable, and the time variable as class variables and uses them and their interactions to regress the response variable.

10. A computer program product for determining an incremental effect model, the computer program product comprising:

a non-transitory computer-readable medium comprising;

an executable portion for causing a computer to determine at least one model variable from a plurality of predictor variables, wherein the model variable is associated with a response variable or a performance variable and a decision variable for a population, wherein the decision variable defines treated observations and control observations in the population;

an executable portion for causing a computer to determine a difference in the response variable between the treated observations and the control observations;

an executable portion for causing a computer to determine a function describing the relationship between the difference and the model variable;

an executable portion for causing a computer to rank order the treated and control observations in the population based on a score, which is a weighted sum of functions of some model variables;

an executable portion for causing a computer to determine a cumulative incremental effect between the response value for the treated and control observations at increasing percentages of ranked observations with the lowest score values;

an executable portion for causing a computer to determine an incremental effect area index based on the cumulative incremental effect as a function of the increasing percentages of ranked observations with the lowest score values; and

an executable portion for causing a computer to iterate through the functions of the model variables until the incremental effect area index is maximized to determine the incremental effect model.

11. The computer program product according to claim 10, wherein the non-transitory computer-readable medium further comprises an executable portion for causing a computer to rank order the model variables.

12. The computer program product according to claim 11, wherein the model variables are rank ordered by a p-value associated with a two-term interaction between the model variable and the decision variable.

13. The computer program product according to claim 10, wherein the non-transitory computer-readable medium further comprises an executable portion for causing a computer to weight the function to define a score, wherein the score is used to rank order the observations.

14. The computer program product according to claim 13, wherein the non-transitory computer-readable medium further comprises an executable portion for causing a computer to create the score by adding the existing score with a product of the weight and the nonlinear function or transformation of the next model variable.

15. The computer-program product according to claim 10, wherein the function describing the relationship between the incremental effect and the model variable, wherein the function is a piecewise linear function when the model variable is a continuous variable.

16. The computer-program product according to claim 10, wherein the non-transitory computer-readable medium further comprises an executable portion to convert continuous predictor variables to categorical predictor variables based on binning.

17. The computer-program product according to claim 10, wherein the difference in the response variable between the treated observations and the control observations is a difference between an average value for the response variable for the treated observations and an average value for the response variable for the control observations within bins which categorize the population based on the values of a model variable.

18. A computer-implemented method of determining an incremental effect model, the method comprising:

determine a set of model variables from a plurality of predictor variables that are decision relevant and optionally time invariant, wherein the predictor variables are associated with a response variable or a performance variable and a decision variable for a population, wherein the decision variable defines treated observations and control observations in the population;

determine a function describing the relationship between an incremental effect and the model variable, wherein the incremental effect is defined as the difference between performance of an individual under treatment and the performance of the individual under no treatment;

rank order the treated and control observations in the population based on a score from low to high, wherein the score is a weighted sum of a set of functions created from the model variables;

determine a cumulative incremental effect at increasing percentages of ranked observations with the lowest score values, based on the combined test and control data;

determine an incremental effect area index based on the cumulative incremental effect as a function of the increasing percentages of ranked observations with the lowest score values; and

iterate through the functions of the model variables for weight search until the incremental effect area index is maximized to determine an incremental effect model;

rank ordering the model variables based on a significance value for a two-term interaction between the decision variable and a predictor variable;

determining a function describing the relationship between the incremental effect and the model variable.

19. The computer-implemented method of claim 18, further comprising determining which individuals to treat based on the incremental effect model.

20. The computer-implemented method of claim 19, wherein the individuals are selected for treatment based on a higher positive response rate predicted by the incremental effect model.