US 12,327,251 B2
Systems, methods, and apparatuses for implementing user customizable risk management tools with statistical modeling and recommendation engine
Michael Manapat, San Francisco, CA (US); Isaac Hepworth, Boulder, CO (US); Tara Seshan, San Francisco, CA (US); and Mike Towber, San Francisco, CA (US)
Assigned to Stripe, Inc., South San Francisco, CA (US)
Filed by Stripe, Inc., South San Francisco, CA (US)
Filed on Mar. 23, 2023, as Appl. No. 18/125,390.
Application 18/125,390 is a continuation of application No. 17/119,069, filed on Dec. 11, 2020, granted, now 11,620,652.
Application 17/119,069 is a continuation of application No. 15/787,426, filed on Oct. 18, 2017, granted, now 10,867,303, issued on Dec. 15, 2020.
Prior Publication US 2023/0230090 A1, Jul. 20, 2023
Int. Cl. G06Q 20/40 (2012.01); G06Q 10/067 (2023.01); G06F 3/0482 (2013.01); G06F 3/04847 (2022.01); G06F 3/0488 (2022.01)
CPC G06Q 20/4016 (2013.01) [G06Q 10/067 (2013.01); G06Q 20/405 (2013.01); G06F 3/0482 (2013.01); G06F 3/04847 (2013.01); G06F 3/0488 (2013.01)] 23 Claims
OG exemplary drawing
 
1. A method for generating score output explanations for transactions performed for a user computer system, the method comprising:
processing, by a server computer system over a communications network, a plurality of transactions on behalf of the user computer system;
evaluating, by the server computer system, the plurality of transactions via a machine learning model by generating a score as an output of the machine learning model for each one of the plurality of transactions before authorization or rejection of each one of the plurality of transactions;
in response to a transaction having the score satisfying a threshold, identifying, by the server computer system, an explanation for a rejection of the transaction predicted by the machine learning model by:
iteratively evaluating, by the server computer system, the transaction using the machine learning model to identify one or more transaction features and one or more corresponding values for the one or more transaction features that provide a greatest contribution to the output of the machine learning model for the transaction, wherein the identified one or more transaction features and the one or more corresponding values form an indicator associated with an authorization or rejection of the transaction;
transmitting, by the server computer system to the user computer system over the communications network, the explanation including a subset of the one or more transaction features and the one or more corresponding values as the indicator causing the user computer system to display the indicator generated for the transaction within a graphical user interface (GUI);
receiving, by the server computer system, a user-defined rule modifying a default behavior of the server computer system;
in response to the transaction satisfying the user-defined rule and having the score that satisfies the threshold, overriding, by the server computer system, a default transaction authorization of the transaction;
permitting, by the server computer system, transmission of the transaction at a sampling rate correlated to the score, wherein the sampling rate comprises:
sampling at a first rate when the score is within a low-score grouping;
sampling at a second rate when the score is within an elevated score grouping, wherein the second rate is lower than the first rate; and
sampling at a third rate when the score is in a high score grouping, wherein the third rate is lower than the second rate;
monitoring, by the server computer system, a determination by an issuer computer system, wherein the determination is whether the transaction is determined to be non-fraudulent or fraudulent after being permitted;
determining, by the server computer system, to what extent the user-defined rule is blocking non-fraudulent transactions based on the determination by the issuer computer system; and
training, by the server computer system, the machine learning model based on the extent to which the user-defined rule is blocking non-fraudulent transactions.