SYSTEMS AND METHODS FOR GENERATING ADVERSE-ACTION REPORTS FOR ADVERSE CREDIT-APPLICATION DETERMINATIONS
Disclosed herein are systems and methods for generating adverse-action reasons for adverse credit-application determinations. In an embodiment, a server calculates a log-odds contribution for each of multiple input variables of a machine-learning model for each of multiple training-data observations. The server identifies a working minimum log-odds contribution and from that a maximum score for each input variable. The server calculates a log-odds contribution and from that an actual score for each input variable for an application data set. The server calculates a score difference between the maximum and actual scores for each input variable. The server outputs an adverse-action report that is associated with the application data set. The adverse-action report includes an indication for each of a predefined number of the input variables having the highest score differences.
1 . A method comprising:
for each of M training-data observations, calculating a respective log-odds contribution for each of N monotonic input variables of a machine-learning model;
for each of the N input variables, identifying a respective working minimum log-odds contribution;
for each of the N input variables, calculating a respective maximum score based on the identified working minimum log-odds contribution for the respective input variable;
receiving an application data set comprising a respective value for each of the N input variables of the machine-learning model;
calculating, based on the machine-learning model and the respective values in the application data set, a respective log-odds contribution for the application data set for each of the N input variables;
for each of the N input variables, calculating a respective actual score based on the respective calculated log-odds contribution for the application data set;
for each of the N input variables, calculating a score difference as the difference between the respective maximum score and the respective actual score; and
outputting an adverse-action report associated with the application data set, the adverse-action report comprising a respective indication for each of a predefined number of the N input variables having the highest calculated score differences.
2 . The method of claim 1 , further comprising generating the machine-learning model based on the M training-data observations
3 . The method of claim 2 , wherein generating the machine-learning model based on the M training-data observations comprises converting any of the N input variables that were previously non-monotonic input variables to being monotonic input variables.
4 . The method of claim 1 , wherein the identified working minimum log-odds contribution for a given input variable is a maximum calculated log-odds contribution of a predefined lowest percent of the calculated log-odds contributions for the given input variable.
5 . The method of claim 1 , wherein the application data set is associated with a credit application for which an adverse determination has been made.
6 . The method of claim 1 , wherein outputting the adverse-action report comprises transmitting the adverse-action report via a data connection to a remote device.
7 . The method of claim 1 , wherein outputting the adverse-action report comprises one or both of storing and updating one or more data files in data storage based on the adverse-action report.
8 . The method of claim 1 , wherein each indication in the adverse-action report comprises one or more of an identification of the corresponding input variable, an adverse-action reason code that is associated with the corresponding input variable, and a definition of an adverse-action reason code that is associated with the corresponding input variable.
9 . The method of claim 1 , wherein the predefined number is between 1 and 10, inclusive.
10 . A system comprising:
a communication interface;
a processor; and
data storage that contains instructions executable by the processor for carrying out a set of functions, the set of functions comprising:
for each of M training-data observations, calculating a respective log-odds contribution for each of N monotonic input variables of a machine-learning model;
for each of the N input variables, identifying a respective working minimum log-odds contribution;
for each of the N input variables, calculating a respective maximum score based on the identified working minimum log-odds contribution for the respective input variable;
receiving an application data set comprising a respective value for each of the N input variables of the machine-learning model;
calculating, based on the machine-learning model and the respective values in the application data set, a respective log-odds contribution for the application data set for each of the N input variables;
for each of the N input variables, calculating a respective actual score based on the respective calculated log-odds contribution for the application data set;
for each of the N input variables, calculating a score difference as the difference between the respective maximum score and the respective actual score; and
outputting an adverse-action report associated with the application data set, the adverse-action report comprising a respective indication for each of a predefined number of the N input variables having the highest calculated score differences.
11 . The system of claim 10 , the set of functions further comprising generating the machine-learning model based on the M training-data observations
12 . The system of claim 11 , wherein generating the machine-learning model based on the M training-data observations comprises converting any of the N input variables that were previously non-monotonic input variables to being monotonic input variables.
13 . The system of claim 10 , wherein the identified working minimum log-odds contribution for a given input variable is a maximum calculated log-odds contribution of a predefined lowest percent of the calculated log-odds contributions for the given input variable.
14 . The system of claim 10 , wherein the application data set is associated with a credit application for which an adverse determination has been made.
15 . The system of claim 10 , wherein outputting the adverse-action report comprises transmitting the adverse-action report via a data connection to a remote device.
16 . The system of claim 10 , wherein outputting the adverse-action report comprises one or both of storing and updating one or more data files in data storage based on the adverse-action report.
17 . The system of claim 10 , wherein each indication in the adverse-action report comprises one or more of an identification of the corresponding input variable, an adverse-action reason code that is associated with the corresponding input variable, and a definition of an adverse-action reason code that is associated with the corresponding input variable.
18 . The system of claim 10 , wherein the predefined number is between 1 and 10, inclusive.
19 . A method comprising:
generating a machine-learning model having N monotonic input variables;
receiving an application data set comprising a respective value for each of the N input variables of the machine-learning model;
calculating, based on the machine-learning model and the respective values in the application data set, a respective log-odds contribution for the application data set for each of the N input variables; and
outputting an adverse-action report associated with the application data set, the adverse-action report comprising a respective indication for each of a predefined number of the N input variables having the highest absolute values of respective calculated log-odds contributions.
20 . The method of claim 19 , wherein outputting the adverse-action report comprises one or more of transmitting the adverse-action report via a data connection to a remote device, storing one or more data files in data storage based on the adverse-action report, and updating one or more data files in data storage based on the adverse-action report.