US 12,437,218 B2
Methods of explaining an individual predictions made by predictive processes and/or predictive models
Gregory Dean Jorstad, Draper, UT (US); and Thomas Patrick Prendergast, Draper, UT (US)
Assigned to Synchrony Bank, Stamford, CT (US)
Filed by Synchrony Bank, Stamford, CT (US)
Filed on Jul. 25, 2023, as Appl. No. 18/358,098.
Application 18/358,098 is a continuation of application No. 17/930,076, filed on Sep. 7, 2022, granted, now 11,810,009.
Application 17/930,076 is a continuation of application No. 16/293,407, filed on Mar. 5, 2019, granted, now 11,475,322, issued on Oct. 18, 2022.
Prior Publication US 2023/0419134 A1, Dec. 28, 2023
Int. Cl. G06N 5/04 (2023.01); G06N 20/00 (2019.01); G06F 3/14 (2006.01)
CPC G06N 5/04 (2013.01) [G06N 20/00 (2019.01); G06F 3/14 (2013.01)] 21 Claims
OG exemplary drawing
 
1. A computer-implemented method, comprising:
obtaining an original dataset, wherein the original dataset is used to train a machine learning model prior to deployment of the machine learning model, and wherein the original dataset includes different sets of values corresponding to different input variables;
generating different sets of sample bins corresponding to the different input variables, wherein the different sets of sample bins are generated according to a probability distribution of the different sets of values;
repeatedly processing the different sets of values from the different sets of sample bins through the machine learning model to generate a set of sample predictions, wherein processing includes using a set of values from a set of sample bins corresponding to an individual input variable and original values corresponding to other input variables from the original dataset, as input to the machine learning model;
identifying a set of influential input variables from the different input variables by comparing the set of sample predictions to an actual prediction, wherein the actual prediction is generated by processing the original dataset through the machine learning model;
receiving a request to generate a prediction from an input record, wherein the input record includes the different input variables and different sets of actual values corresponding to the different input variables;
updating the input record by modifying one or more actual values corresponding to the set of influential input variables, wherein the one or more actual values are modified by selecting a subset of sample values from a set of sample bins corresponding to the set of influential input variables; and
processing the updated input record through the machine learning model to generate the prediction.