Granted Patent B2

US 12,265,918 · App. 18/503,312 · Granted Apr 1, 2025

Inventors: Douglas C. Merrill (Burbank, CA); Armen Donigian (Burbank, CA); Eran Dvir (Burbank, CA); Sean Javad Kamkar (Burbank, CA); Evan George Kriminger (Burbank, CA); Vishwaesh Rajiv (Burbank, CA); Michael Edward Ruberry (Burbank, CA); Ozan Sayin (Burbank, CA); Yachen Yan (Burbank, CA); Derek Wilcox (Burbank, CA); John Candido (Burbank, CA); Benjamin Anthony Solecki (Burbank, CA); Jiahuan He (Burbank, CA); Jerome Louis Budzik (Burbank, CA); John J. Beahan, Jr. (Burbank, CA); John Wickens Lamb Merrill (Burbank, CA); Esfandiar Alizadeh (Burbank, CA); Liubo Li (Burbank, CA); Carlos Alberta Huertas Villegas (Burbank, CA); Feng Li (Burbank, CA); Randolph Paul Sinnott, Jr. (Burbank, CA)

Assignee: ZestFinance, Inc.

G06N5/022G06F16/908G06F18/2115G06F40/44G06N5/045G06N20/00G06Q10/0633

Abstract

Claims (28)

1. A method implemented by a modeling system and comprising:

detecting one or more unexpected scores during validation of a model before deploying the model in a production environment, comprising:

comparing a validation distribution of scores with a reference distribution of scores, wherein the validation distribution of scores is generated by the model for a validation input data set and the reference distribution of scores is generated by the model for a test input data set;

determining whether an alert condition is satisfied based on a shift of a plurality of median model score values in the validation distribution of scores, wherein the shift is identified based on the comparison; and

providing an alert when the alert condition is determined to be satisfied, wherein the alert indicates detection of the one or more unexpected scores.

2. The method of claim 1 , wherein the reference distribution of scores comprises, for each of a plurality of score values, a vector comprising a distribution value indicating a portion of the scores in the reference distribution of scores that have one of the score values.

3. The method of claim 1 , further comprising generating a population stability index (PSI), or performing a Kolmogorov-Smirnov test, using the validation distribution of scores and the reference distribution of scores to determine whether to provide the alert.

4. The method of claim 1 , further comprising storing in a structured database model documentation generated based on metadata collected during processing of a workflow associated with the model, wherein the model documentation comprises one or more of fair lending documentation, a model comparison, a sensitivity analysis, feature importance data, unexpected input data information, unexpected score information, or an adverse action reason code mapping.

5. A modeling system, comprising memory having instructions stored thereon and one or more processors coupled to the memory and configured to execute the stored instructions to:

detect an unexpected score during execution of a model in a production environment, comprising:

compare a production distribution of first scores, which are generated by the model during the execution of the model in the production environment during a first time period, with a reference distribution of second scores, which are generated by the model during the execution of the model in the production environment during a second time period prior to the first time period;

determine whether to provide an alert based on a shift of a plurality of median model score values in the production distribution of first scores, wherein the shift is identified based on the comparison; and

selectively provide the alert to an external system based on the determination, wherein the alert indicates detection of the unexpected score.

6. The modeling system of claim 5 , wherein the reference distribution comprises, for each of a plurality of score values, a vector comprising a distribution value indicating a portion of the second scores that have one of the score values.

7. The modeling system of claim 5 , wherein the one or more processors are further configured to execute the stored instructions to generate a population stability index (PSI), or performing a Kolmogorov-Smirnov test, using the production distribution and the reference distribution to determine whether to provide the alert to the external system.

8. The modeling system of claim 5 , wherein the one or more processors are further configured to execute the stored instructions to store in a structured database model documentation generated based on metadata collected during processing of a workflow associated with the model.

9. The modeling system of claim 8 , wherein the model documentation comprises one or more of fair lending documentation, a model comparison, a sensitivity analysis, feature importance data, unexpected input data information, unexpected score information, or an adverse action reason code mapping.

10. A non-transitory computer readable medium having stored thereon instructions comprising executable code that, when executed by one or more processors, causes the one or more processors to:

for each of a plurality of score, generate a reference distribution of score values, wherein the reference distribution of score values comprises, for each of the score values, a vector comprising a distribution value indicating a percentage of the plurality of reference input data sets having the score value; and

detect an unexpected score during execution of the model, comprising, for each score of one or more production input data sets used by the model:

compare a production distribution of score values for the score with the reference distribution of score values for the score wherein the production distribution of score values comprises a plurality of median model scores values;

identify a shift of at least one of the median model score values based on the comparison; and

provide an alert to an external system, wherein the alert indicates detection of the unexpected score.

11. The non-transitory computer readable medium of claim 10 , wherein the production distribution of score values is a distribution of scores within a first subset of the production input data sets and the reference distribution of scores is another distribution of scores within a second subset of the production input data sets.

12. The non-transitory computer readable medium of claim 10 , wherein the reference distribution of score values is within each of a plurality of data sets used to validate or train the model.

13. The non-transitory computer readable medium of claim 10 , wherein the executable code, when executed by the one or more processors, further causes the one or more processors to generate a population stability index (PSI), or performing a Kolmogorov-Smirnov test, using the production distribution of score values and the reference distribution of score values to determine whether to provide the alert to the external system.

14. The non-transitory computer readable medium of claim 10 , wherein the executable code, when executed by the one or more processors, further causes the one or more processors to store in a structured database model documentation generated based on metadata collected during processing of a workflow associated with the model.

15. The non-transitory computer readable medium of claim 14 , wherein the model documentation comprises one or more of fair lending documentation, a model comparison, a sensitivity analysis, feature importance data, unexpected input data information, unexpected score information, or an adverse action reason code mapping.

Assignments (1)

Continuity (3)

References Cited (318)