Method and system for improving machine learning operation by reducing machine learning bias
A network operation system and method accesses a training dataset for a network operation predictive model including historical network operation records and historical decision records, generates an inferred protected class dataset by executing a protected class demographic model, executes an algorithmic bias model using as input the historical decision records and the inferred protected class dataset to generate one or more fairness metrics, executes, based on the fairness metrics, a bias adjustment model using as input the historical decision records and the inferred protected class dataset to generate an adjusted training dataset, trains the network operation predictive model using as input the adjusted training dataset, receives an electronic request for a network operation, executes the network operation predictive model using as input at least one attribute of the electronic request for the network operation, and executes the network operation based on a prediction of the network operation predictive model.
1. A method of improving efficiency of a machine learning model by reducing bias in the machine learning model via iterative training, the method comprising:
accessing, by a processor, a training dataset for a network operation predictive model comprising a plurality of historical network operation records and a plurality of historical decision records each representing a historical decision whether to accept a respective historical network operation;
generating, by the processor, an inferred protected class dataset by executing a protected class demographic model using as input the plurality of historical network operation records, wherein the inferred protected class dataset identifies predicted demographic groups for the plurality of historical network operation records;
executing, by the processor, an algorithmic bias model using as input the plurality of historical decision records and the inferred protected class dataset to generate one or more first fairness metrics for the plurality of historical decision records;
executing, by the processor, based on the one or more first fairness metrics, a bias adjustment model using as input the plurality of historical decision records and the inferred protected class dataset to generate an adjusted training dataset, wherein the adjusted training dataset includes only features correlated with target variables;
training, by the processor, the network operation predictive model by executing the network operation predictive model using as input the adjusted training dataset;
executing, by the processor, the network operation predictive model using as input a plurality of network operation records to generate a plurality of decision records;
executing, by the processor, the algorithmic bias model using as input the plurality of decision records to generate one or more second fairness metrics for the plurality of decision records generated by the network operation predictive model;
executing, by the processor, based on the one or more second fairness metrics, the bias adjustment model to adjust the network operation predictive model to increase the one or more second fairness metrics generated by the algorithmic bias model;
receiving, by the processor, an electronic request for a network operation;
executing, by the processor, the network operation predictive model using as input at least one attribute of the electronic request for the network operation; and
executing, by the processor, the network operation based on a prediction of the network operation predictive model.
2. The method of claim 1 , wherein generating the adjusted training dataset includes removing discriminatory features from the plurality of historical network operation records.
3. The method of claim 2 , wherein removing the discriminatory features from the training dataset includes screening features to include only features that correlate with target variables.
4. The method of claim 1 , further comprising training the protected class demographic model by comparing an output of the protected class demographic model to actual demographic information.
5. The method of claim 1 , further comprising removing, from the training dataset, overt identifiers of demographic information prior to generating the inferred protected class dataset.
6. The method of claim 1 , wherein the network operation predictive model is configured to output a decision whether to extend credit to a user.
7. The method of claim 1 , wherein the one or more fairness metrics include a fairness metric for a credit score for each of the historical network operation records.
8. The method of claim 1 , further comprising:
generating a second set of decision records by executing the network operation predictive model using as input a second set of network operation records;
generating a second inferred protected class dataset by executing the protected class demographic model using as input the second set of network operation records;
executing, by the processor, the algorithmic bias model on the second set of decision records and the second inferred protected class dataset to generate one or more second fairness metrics for the second set of decision records;
executing, by the processor, based on the one or more second fairness metrics, the bias adjustment model using as input the second set of decision records and the second inferred protected class dataset to generate a second adjusted training dataset; and
training the network operation predictive model using as input the second adjusted training dataset.
9. The method of claim 8 , wherein the second adjusted training dataset includes one or more decision records of the second set of decision records having a fairness metric below a predetermined threshold.
10. The method of claim 8 , wherein generating the second adjusted training dataset includes adjusting, based on the one or more second fairness metrics, one or more decision records of the second set of decision records.
11. A system, comprising:
a network operation predictive model;
a non-transitory machine-readable memory that stores a training dataset for the network operation predictive model comprised of a plurality of historical network operation records and a plurality of historical decision records each representing a decision whether to accept a respective historical network operation; and
a processor, wherein the processor in communication with the network operation predictive model and the non-transitory, machine-readable memory executes a set of instructions instructing the processor to:
generate an inferred protected class dataset by executing the protected class demographic model using as input the plurality of historical network operation records, wherein the inferred protected class dataset identifies predicted demographic groups for the plurality of historical network operation records;
execute an algorithmic bias model using as input the plurality of historical decision records and the inferred protected class dataset to generate one or more first fairness metrics for the plurality of historical decision records;
execute, based on the one or more first fairness metrics, a bias adjustment model using as input the plurality of historical decision records and the inferred protected class dataset to generate an adjusted training dataset, wherein the adjusted training dataset includes only features correlated with target variables;
train the network operation predictive model by executing the network operation predictive model using as input the adjusted training dataset;
execute the network operation predictive model using as input a plurality of network operation records to generate a plurality of decision records;
execute the algorithmic bias model using as input the plurality of decision records to generate one or more second fairness metrics for the plurality of decision records generated by the network operation predictive model;
execute, based on the one or more second fairness metrics, the bias adjustment model to adjust the network operation predictive model to increase the one or more second fairness metrics generated by the algorithmic bias model;
receive an electronic request for a network operation;
execute the network operation predictive model using as input at least one attribute of the electronic request for the network operation; and
execute the network operation based on a prediction of the network operation predictive model.
12. The system of claim 11 , wherein the instructions instruct the processor to generate the adjusted training dataset includes removing discriminatory features from the plurality of historical network operation records.
13. The system of claim 12 , wherein the instructions instruct the processor to remove the discriminatory features from the training dataset includes screening features to include only features that correlate with target variables.
14. The system of claim 11 , wherein the instructions instruct the processor to train the protected class demographic model by comparing an output of the protected class demographic model to actual demographic information.
15. The system of claim 11 , wherein the instructions instruct the processor to remove, from the training dataset, overt identifiers of demographic information prior to generating the inferred protected class dataset.
16. The system of claim 11 , wherein the network operation predictive model is configured to output a decision whether to extend credit to a user.
17. The system of claim 11 , wherein the one or more fairness metrics include a fairness metric for a credit score for each of the historical network operation records.
18. The system of claim 11 , wherein the instructions instruct the processor to:
generate a second set of decision records by executing the network operation predictive model using as input a second set of network operation records;
generate a second inferred protected class dataset by executing the protected class demographic model using as input the second set of network operation records;
execute the algorithmic bias model on the second set of decision records and the second inferred protected class dataset to generate one or more second fairness metrics for the second set of decision records;
execute based on the one or more second fairness metrics, the bias adjustment model using as input the second set of decision records and the second inferred protected class dataset to generate a second adjusted training dataset; and
train the network operation predictive model using as input the second adjusted training dataset.
19. The system of claim 18 , wherein the second adjusted training dataset includes one or more decision records of the second set of decision records having a fairness metric below a predetermined threshold.
20. The system of claim 18 , wherein the instructions instruct the processor to generate the second adjusted training dataset by adjusting, based on the one or more second fairness metrics, one or more decision records of the second set of decision records.