Process database entries to provide predictions of future data values

• US 6,725,210 B1
• Filed: 11/15/2000
• Issued: 04/20/2004
• Est. Priority Date: 11/20/1999
• Status: Active Grant

• Current Assignee: Teradata US, Inc. (Teradata Corporation)
• Original Assignee: NCR Corporation
• Inventors: Tan, Say Beng, Key, Jane
• Primary Examiner(s): Khatri, Anil
• Assistant Examiner(s): Holmes, Michael B.
• Application Number:
  US09/713,368
• Time in Patent Office: 1,252 Days
• Field of Search: 706/45-60
• US Class Current: 706/45
• CPC Class Codes: G06F 17/15 Correlation function computation including computation of convolution operations arithmetic circuits for sum of products per se, e.g. multiply-accumulators G06F7/5443; digital filters, e.g. FIR, IIR, adaptive filters H03H17/00
• CPC Class Codes: G06F 17/18 for evaluating statistical data , e.g. average values, frequency distributions, probability functions, regression analysis forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04
• CPC Class Codes: G06Q 10/06 Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling

Abstract

The present invention relates to a method and apparatus for processing entries stored in an electronic database where each entry comprises a succession of data values of correlated variables. The entries are processed in order to provide predictions of future data values of the correlated variables. The invention has particular application to processing data entries that relate to customers of a bank or retail business so as to predict future data values of attributes of the customers. A prior distribution over the parameters of a first of the variables is accessed from the database and data values of the first variable are used to derive a posterior distribution of the parameters of the first variable. A prior distribution over the parameters of a second of the variables is accessed from the database, the second of the variables being correlated with the first. Data values of the second variable are accessed to derive a posterior distribution of the parameters of the second variable. A statistical sampler samples the parameters of each posterior distribution to provide estimates of the parameters of the posterior distributions. Predictive distributions of the first and second variables are computed from the estimates. Predicted data values for the first and second correlated variables are then formulated from the predictive distributions and known values for the variables.

Claims

1. A method of operating a data processing apparatus to process variables stored in an electronic database so as to provide prediction of future values of the variables, the method comprising the steps of:

accessing a prior distribution over the parameters of a first of the variables;

accessing data values of the first variable to derive a posterior distribution of the parameters of the first variable;

accessing a prior distribution over the parameters of the second of the variables, the second of the variables being correlated with the first;

accessing data values of the second variable to derive a posterior distribution of the parameters of the second variable;

taking statistical samples of the parameters of each posterior distribution to provide estimates of the parameters of the posterior distributions of each of the first and the second variables;

computing a predictive distribution of the first variable from said estimates, both the estimate of the parameters of the posterior distribution of the first variable and the estimate of the parameters of the posterior distribution of the second variable being used in computing the predictive distribution of the first variable;

computing a predictive distribution of the second variable from the said estimates, both the estimate of the parameters of the posterior distribution of the first variable and the estimate of the parameters of the posterior distribution of the second variable being used in computing the predictive distribution of the second variable; and

formulating predicted data values for the first and second correlated variables from said predictive distributions and known values for the variables.

2. A method as claimed in claim 1, wherein the first variable is a continuous variable and the prior distribution over the parameters of the first variable is a normal distribution.

3. A method as claimed in claim 1, wherein the second variable is a discrete variable and the prior distribution over the parameters of the second variable is a Poisson distribution.

4. A method as claimed in claim 1, wherein the first variable comprises a personal attribute of individuals represented in the database and the second variable concerns transactions made by the individuals and the method comprises the further step of calculating profitability from the predicted data values of the first and second correlated variables.

11. The method of claim 1, wherein the first and second variables relate to attributes associated with a customer of a commercial establishment.

12. The method of claim 11, wherein the attributes relate to account characteristics of a bank customer.

13. The method of claim 12, wherein the first and second variables relate to the profitability of the customer'"'"'s activity to the establishment.

14. The method of claim 13, wherein at least one of the first and second variables relates to transactions undertaken by the customer.

5. A method of operating a data processing apparatus to process variables stored in an electronic database so as to provide prediction of future values of the variables, the method comprising the steps of:

accessing a prior distribution over the parameters of a first plurality of the variables, the first plurality of the variables, the first variables being continuous variables;

accessing data values of the first variables to derive a posterior distribution of the parameters of the first variables;

accessing a prior distribution over the parameters of a second plurality of the variables, the second variables being discrete variables correlated with the first variables;

accessing data values of the second variables to derive a posterior distribution of the parameters of the second variables;

taking statistical samples of the parameters of each posterior distribution to provide estimates of the parameters of the posterior distributions of each of the first and the second plurality of variables;

computing a predictive distribution of the first variables from said estimates, both the estimate of the parameters of the posterior distribution of the first variables and the estimate of the parameters of the posterior distribution of the second variables being used in computing the predictive distribution of the first variables;

computing a predictive distribution of the second variables from said estimates, both the estimate of the parameters of the posterior distribution of the first variables and the estimate of the parameters of the posterior distribution of the second variables being used in computing the predictive distribution of the second variables; and

formulating predicted data values for the first and second correlated variables from said predictive distributions and known values for the variables.

15. The method of claim 5, wherein the first and second pluralities of variables relate to attributes associated with a customer of a commercial establishment.

16. The method of claim 15, wherein the attributes relate to account characteristics of a bank customer.

17. The method of claim 16, wherein the first and second pluralities of variables relate to the profitability of the customer'"'"'s activity to the establishment.

18. The method of claim 17, wherein at least one variable of the first and second pluralities of variables relates to transactions undertaken by the customer.

6. Data processing apparatus to process variables stored in an electronic database so as to provide prediction of future values of the variables, the apparatus comprising:

means for accessing a prior distribution over the parameters of a first of the variables;

means for accessing data values of the first variable to derive a posterior distribution of the parameters of the first variable;

means for accessing a prior distribution over the parameters of a second of the variables, the second of the variables being correlated with the first;

means for accessing data values of the second variable to derive a posterior distribution of the parameters of the second variable;

a statistical sampler of the parameters of each posterior distribution for providing estimates of the parameters of the posterior distributions of each of the first and the second variables;

computing means for computing a predictive distribution of the first variable and a predictive distribution of the second variable from the said estimates, the computing means employing both the estimate of the parameters of the posterior distribution of the first variable and the estimate of the parameters of the posterior distribution of the second variable to compute the predictive distribution of the first variable, the computing means further employing both the estimate of the parameters of the posterior distribution of the first variable and the estimate of the parameters of the posterior distribution of the second variable to compute the predictive distribution of the second variable; and

a predictor for predicting data values for the first and second correlated variables from said predictive distributions and known values for the variables.

7. Apparatus as claimed in claim 6, wherein the first variable is a discrete variable and the prior distribution over the parameters of the first variable is a Poisson distribution.

8. Apparatus as claimed in claim 6, wherein the second variable is a discrete variable and the prior distribution over the parameters of the second variable is a Poisson distribution.

9. Apparatus as claimed in claim 6, wherein the first variable comprises a personal attribute of individuals represented in the database and the second variable concerns transactions made by the individuals, the apparatus further comprising profitability calculating means for calculating profitability from the predicted data values of the first and second correlated variables.

10. Data processing apparatus to process variables stored in an electronic database so as to provide prediction of future values of the variables, the apparatus comprising:

means for accessing a prior distribution over the parameters of a first plurality of the variables, the first variables being continuous variables;

means for accessing data values of the first variables to derive posterior distributions of the parameters of the first variables;

means for accessing prior distributions over the parameters of a second plurality of the variables, the second variables being discrete variables correlated with the first;

means for accessing data values of the second variables to derive posterior distributions of the parameters of the second variables;

a statistical sampler for sampling the parameters of each posterior distribution for providing estimates of the parameters of the posterior distributions of each of the first and the second plurality of variables;

computing means for computing a predictive distribution of the first variables and a predictive distribution of the second variables from said estimates, the computing means employing both the estimate of the parameters of the Posterior distribution of the first variables and the estimate of the parameters of the posterior distribution of the second variables to compute the predictive distribution of the first variables, the computing means further employing both the estimate of the parameters of the Posterior distribution of the first variables and the estimate of the parameters of the posterior distribution of the second variables to compute the predictive distribution of the second variables; and

a predictor for predicting data values for the first and second correlated variables from said predictive distributions and known values for the variables.