Method, system and program product for determining a time for retraining a data mining model

1. A method for determining a time for retraining a data mining model, comprising:

performing the following steps by the performance of a computer processor:

calculating multivariate statistics of a training model during a training phase;

storing said multivariate statistics in the data mining model;

evaluating reliability of the data mining model based on said multivariate statistics; and

deciding to retrain the data mining model based on an arbitrary measure of one or more statistical parameters, including an F-test statistical analysis.

2. The method according to claim 1 further comprising the step of segmenting said multivariate statistics by at least one input field.

3. The method according to claim 1 further comprising the steps of calculating univariate statistics of said training model during said training phase and storing said univariate statistics in the data mining model.

4. The method according to claim 1 further comprising the step of monitoring at least one distribution parameter of application data as a function of time during a deployment phase.

5. The method according to claim 4 wherein said step of evaluating reliability of the data mining model is further based on said at least one distribution parameter.

6. The method according to claim 4 further comprising the step of monitoring the data mining model by using said application data.

7. The method according to claim 4 wherein said step of evaluating reliability of the data mining model comprises the step of using a subset of training data corresponding to a segment concurrently present in the application data to monitor application of the data mining model.

8. The method according to claim 4 further comprising the steps of deploying the data mining model to said application data and performing a statistical analysis of said application data.

9. The method according to claim 8 further comprising the steps of merging and aggregating statistics retrieved from the data mining model to match a particular data subset existing in said application data.

10. The method according to claim 8 wherein said step of performing a statistical analysis comprises the step of locating segments corresponding to an application data distribution.

11. The method according to claim 1 wherein said multivariate statistics comprise bivariate statistics.

12. The method according to claim 1 wherein said statistical parameter comprises at least one of a maximum change over all parameters and a weighted sum of all differences between parameters in application data set statistics and in a corresponding part of multivariate statistics of the data mining model.

13. A data processing system comprising:

a memory element including application data, wherein said application data are assigned to input fields, where at least one distribution parameter of the application data is monitored as a function of time; and

a processor for application of a data mining model to said application data,

wherein the processor calculates multivariate statistics of a training model during a training phase, stores the multivariate statistics in the data mining model, and evaluates reliability of the data mining model based on the multivariate statistics; and

wherein the processor retrains the data mining model based on an arbitrary measure of one or more statistical parameters, including an F-test statistical analysis.

14. The system according to claim 13 wherein said memory element is coupled to said processor.

15. The system according to claim 13 wherein said memory element comprises multivariate statistics.

16. A program product comprising a computer useable medium including a computer readable program, wherein the computer readable program when executed on a computer causes the computer to perform the steps of:

calculating multivariate statistics of a training model during a training phase;

storing said multivariate statistics in a data mining model; and

evaluating reliability of said data mining model based on said multivariate statistics; and

deciding to retrain the data mining model based on an arbitrary measure of one or more statistical parameters including an F-test statistical analysis.

17. The program product according to claim 16 further comprising the step of monitoring at least one distribution parameter of application data during deployment of said data mining model to application data.

18. The program product according to claim 16 further comprising the step of evaluating reliability of said data mining model by using statistics retrieved from said data mining model to match a particular data subset in application data.