US 12,321,826 B2
System and method for utilizing grouped partial dependence plots and game-theoretic concepts and their extensions in the generation of adverse action reason codes
Alexey Miroshnikov, Evanston, IL (US); Konstandinos Kotsiopoulos, Easthampton, MA (US); Arjun Ravi Kannan, Buffalo Grove, IL (US); Raghu Kulkarni, Buffalo Grove, IL (US); and Steven Dickerson, Deerfield, IL (US)
Assigned to Discover Financial Services, Riverwoods, IL (US)
Filed by Discover Financial Services, Riverwoods, IL (US)
Filed on May 17, 2021, as Appl. No. 17/322,828.
Application 17/322,828 is a continuation in part of application No. 16/868,019, filed on May 6, 2020, granted, now 12,050,975.
Prior Publication US 2021/0383275 A1, Dec. 9, 2021
Int. Cl. G06N 20/00 (2019.01); G06F 17/16 (2006.01); G06F 18/2413 (2023.01); G06N 3/084 (2023.01); G06N 5/045 (2023.01); G06V 10/774 (2022.01)
CPC G06N 20/00 (2019.01) [G06F 17/16 (2013.01); G06F 18/24137 (2023.01); G06N 3/084 (2013.01); G06N 5/045 (2013.01); G06V 10/7753 (2022.01)] 22 Claims
OG exemplary drawing
 
1. A method, comprising:
training a machine learning (ML) model by carrying out a machine learning process on a training dataset comprising a set of input vectors and a corresponding set of output values, wherein the trained machine learning model is configured to (i) receive an input vector comprising respective values for a given set of input variables and (ii) based on an evaluation of the received input vector, generate a prediction;
defining a plurality of variable groups based on an evaluation of dependencies between input variables in the given set of input variables of the ML model, wherein (i) each variable group includes at least one input variable from the given set of input variables and (ii) each of at least a subset of the variable groups includes two or more input variables from the given set of input variables that are determined to have a threshold level of dependence with one another;
after training the machine learning model and dividing the given set of input variables into the plurality of variable groups:
receiving a given input vector comprising respective values for the given set of input variables;
processing, by the ML model, the given input vector to generate an output vector a given prediction corresponding to the given input vector;
evaluating the given input vector and the given prediction using (i) the defined plurality of variable groups and (ii) at least one model interpretability technique that serves to explain the ML model by quantifying contributions of the given set of input variables to predictions generated by the ML model;
based on the evaluating, generating a respective score for each variable group in the defined plurality of variable groups, wherein the respective score for each variable group quantifies a respective contribution of the variable group to the given prediction; and
based on the respective scores determined for the defined plurality of variable groups, identifying one or more input variables of the ML model that contributed most to the given prediction.