Metric specific machine learning model improvement through metric specific outlier removal

1. A computer-implemented method for machine learning model training outliers, comprising:

providing a cloud service to update a machine learning model for a business unit such that a key performance indicator (KPI) of the business unit meets a predetermined KPI goal, by:

determining a numerically quantified metric improvement goal based on a standard unit value that represents a correlation between the KPI and a metric of the machine learning model;

generating a plurality of scored transactions comprising a determined metric value for each of a corresponding plurality of transactions generated for the machine learning model, wherein the determined metric value is associated with the numerically quantified metric improvement goal; identifying a plurality of preliminary outliers from the plurality of scored transactions based on a derived cut-off value for the determined metric value;

deleting the plurality of preliminary outliers;

identifying a plurality of training outliers from a plurality of remaining transactions, wherein the remaining transactions comprise the transactions remaining after deleting the preliminary outliers, and wherein training the machine learning model with the training outliers influences the numerically quantified metric improvement goal;

generating re-training data by re-labeling the plurality of training outliers with a plurality of predetermined labels corresponding to the plurality of training outliers; and

re-training a classifier of the machine learning model with the re-training data, wherein the re-trained classifier meets the numerically quantified metric improvement goal; and

providing the updated machine learning model to meet the KPI.

2. The computer-implemented method of claim 1 , further comprising calculating the numerically quantified metric improvement goal based on a relative goal.

3. The computer-implemented method of claim 2 , wherein the numerically quantified metric improvement goal is equal to (1+rg) X mv, wherein rg represents a relative goal expressed as a digital value of a percentage, and wherein my represents a metric value.

4. The computer-implemented method of claim 2 , wherein the relative goal comprises a relative KPI goal.

5. The computer-implement method of claim 1 , wherein the derived cut-off value comprises a metric value of the transactions.

6. The computer-implemented method of claim 1 , wherein the method further comprises deleting the preliminary outliers by:

aggregating the transactions into batches of predetermined size based on a predetermined criteria;

sorting the batches based on the determined metric value;

calculating the cut-off value; and

identifying the preliminary outliers based on the cut-off value.

7. The computer-implemented method of claim 1 , wherein identifying the plurality of training outliers comprises repeating the following until a metric value of the remaining transactions meets the numerically quantified metric improvement goal:

determining an index based on a relatively lowest metric value of the remaining transactions;

identifying the plurality of training outliers based on the index; and

removing the plurality of training outliers from the transactions.

8. A computer program product comprising program instructions stored on a computer readable storage medium, the program instructions executable by a processor to cause the processor to perform a method comprising:

providing a cloud service to update a machine learning model for a business unit such that a key performance indicator (KPI) of the business unit meets a predetermined KPI goal, by:

determining a numerically quantified metric improvement goal based on a standard unit value that represents a correlation between the KPI and a metric of the machine learning model;

generating a plurality of scored transactions comprising a determined metric value for each of a corresponding plurality of transactions generated for the machine learning model, wherein the determined metric value is associated with the numerically quantified metric improvement goal; identifying a plurality of preliminary outliers from the plurality of scored transactions based on a derived cut-off value for the determined metric value;

deleting the plurality of preliminary outliers;

identifying a plurality of training outliers from a plurality of remaining transactions, wherein the remaining transactions comprise the transactions remaining after deleting the preliminary outliers, and wherein training the machine learning model with the training outliers influences the numerically quantified metric improvement goal;

generating re-training data by re-labeling the plurality of training outliers with a plurality of predetermined labels corresponding to the plurality of training outliers; and

re-training a classifier of the machine learning model with the re-training data, wherein the re-trained classifier meets the numerically quantified metric improvement goal; and

providing the updated machine learning model to meet the KPI.

9. The computer program product of claim 8 , the method further comprising calculating the numerically quantified metric improvement goal based on a relative goal.

10. The computer program product of claim 9 , wherein the numerically quantified metric improvement goal is equal to (1+rg) X mv, wherein rg represents a relative goal expressed as a digital value of a percentage, and wherein my represents a metric value.

11. The computer program product of claim 9 , wherein the relative goal comprises a relative KPI goal.

12. The computer program product of claim 8 , wherein the derived cut-off value comprises a metric value of the transactions.

13. The computer program product of claim 8 , wherein the method further comprises deleting the preliminary outliers by:

aggregating the transactions into batches of predetermined size based on a predetermined criteria;

sorting the batches based on the determined metric value;

calculating the cut-off value; and

identifying the preliminary outliers based on the cut-off value.

14. The computer program product of claim 8 , wherein identifying the plurality of training outliers comprises repeating the following until a metric value of the remaining transactions meets the numerically quantified metric improvement goal:

determining an index based on a relatively lowest metric value of the remaining transactions;

identifying the plurality of training outliers based on the index; and

removing the plurality of training outliers from the transactions.

15. A system comprising:

a computer processing circuit; and

a computer-readable storage medium storing instructions, which, when executed by the computer processing circuit, are configured to cause the computer processing circuit to perform a method comprising:

providing a cloud service to update a machine learning model for a business unit such that a key performance indicator (KPI) of the business unit meets a predetermined KPI goal, by:

determining a numerically quantified metric improvement goal based on a standard unit value that represents a correlation between the KPI and a metric of the machine learning model;

generating a plurality of scored transactions comprising a determined metric value for each of a corresponding plurality of transactions generated for the machine learning model, wherein the determined metric value is associated with the numerically quantified metric improvement goal; identifying a plurality of preliminary outliers from the plurality of scored transactions based on a derived cut-off value for the determined metric value;

deleting the plurality of preliminary outliers;

identifying a plurality of training outliers from a plurality of remaining transactions, wherein the remaining transactions comprise the transactions remaining after deleting the preliminary outliers, and wherein training the machine learning model with the training outliers influences the numerically quantified metric improvement goal;

generating re-training data by re-labeling the plurality of training outliers with a plurality of predetermined labels corresponding to the plurality of training outliers; and

re-training a classifier of the machine learning model with the re-training data, wherein the re-trained classifier meets the numerically quantified metric improvement goal; and

providing the updated machine learning model to meet the KPI.

16. The system of claim 15 , the method further comprising calculating the numerically quantified metric improvement goal based on a relative goal, and wherein the numerically quantified metric improvement goal is equal to (1+rg) X mv, wherein rg represents a relative goal expressed as a digital value of a percentage, and wherein my represents a metric value.

17. The system of claim 15 , wherein the method further comprises deleting the preliminary outliers by:

aggregating the transactions into batches of predetermined size based on a predetermined criteria;

sorting the batches based on the determined metric value;

calculating the cut-off value; and

identifying the preliminary outliers based on the cut-off value.

18. The system of claim 15 , wherein identifying the plurality of training outliers comprises repeating the following until a metric value of the remaining transactions meets the numerically quantified metric improvement goal:

determining an index based on a relatively lowest metric value of the remaining transactions;

identifying the plurality of training outliers based on the index; and

removing the plurality of training outliers from the transactions.