Optimizing neural networks for risk assessment

Patent metadata

Field	Value
Publication number	US-10963791-B2
Application number	US-202017019098-A
Country	US
Kind code	B2
Filing date	Sep 11, 2020
Priority date	Mar 27, 2015
Publication date	Mar 30, 2021
Grant date	Mar 30, 2021

Abstract

Certain embodiments involve generating or optimizing a neural network for risk assessment. The neural network can be generated using a relationship between various predictor variables and an outcome (e.g., a condition's presence or absence). The neural network can be used to determine a relationship between each of the predictor variables and a risk indicator. The neural network can be optimized by iteratively adjusting the neural network such that a monotonic relationship exists between each of the predictor variables and the risk indicator. The optimized neural network can be used both for accurately determining risk indicators using predictor variables and determining adverse action codes for the predictor variables, which indicate an effect or an amount of impact that a given predictor variable has on the risk indicator. The neural network can be used to generate adverse action codes upon which consumer behavior can be modified to improve the risk indicator score.

First claim

What is claimed is:

1. A non-transitory computer-readable medium having program code that is executable by a processing device to perform operations, the operations comprising:
retrieving an input set of predictor values for a set of predictor variables, wherein the input set of predictor values corresponds to a target entity;
determining an output risk indicator by applying a neural network to the input set of predictor values, wherein the neural network is trained to compute values of a risk indicator from the set of predictor variables, and a monotonic relationship exists between each predictor variable and the risk indicator as computed by the neural network;
determining, based on the neural network, impacts of the set of predictor variables, respectively, on the output risk indicator;
generating, based on the neural network, an output set of explanatory codes based on the determined impacts of the set of predictor variables;
generating an electronic communication that includes the output risk indicator and the output set of explanatory codes; and
causing the electronic communication to be sent to a user device for display of the output risk indicator and the output set of explanatory codes at the user device.

2. The non-transitory computer-readable medium of claim 1 , wherein determining the impacts of the set of predictor variables on the output risk indicator comprises:
for each predictor variable in the set of predictor variables,
determining a difference between (1) a maximum output value of the neural network generated using an optimized value of the predictor variable and (2) an output value of the neural network generated using an input value of the predictor variable from the input set of predictor values.

3. The non-transitory computer-readable medium of claim 2 , wherein the optimized value of the predictor variable is an endpoint of a domain of the predictor variable.

4. The non-transitory computer-readable medium of claim 1 , wherein the operations further comprise determining ranks of two or more predictor variables in the set of predictor variables based on the impacts of the predictor variables on the output risk indicator, wherein the output set of explanatory codes are generated based on the ranks of the two or more predictor variables, and the output set of explanatory codes includes data describing a contribution of at least one of the two or more predictor variables to the output risk indicator.

5. The non-transitory computer-readable medium of claim 1 , wherein:
the set of predictor variables having a first predictor variable and a second predictor variable;
determining the impacts of the set of predictor variables on the output risk indicator comprises:
determining a set of maximum predictor values including a first maximum value of the first predictor variable and a second maximum value of the second predictor variable, wherein applying the neural network to the set of maximum predictor values generates a maximum output value of the neural network,
computing the impact of the first predictor variable on the output risk indicator by (i) modifying the set of maximum predictor values to replace only the first maximum value with a first input value of the first predictor variable from the input set of predictor values and (ii) computing a first difference between the maximum output value and a first modified output value that is generated by applying the neural network to the modified set of maximum predictor values,
computing the impact of the second predictor variable to the output risk indicator by (i) modifying the set of maximum predictor values to replace only the second maximum value with a second input value of the second predictor variable from the input set of predictor values and (ii) computing a second difference between the maximum output value and a second modified output value that is generated by applying the neural network to the modified set of maximum predictor values; and

generating the output set of explanatory codes based on the determined impacts of the set of predictor variables comprises:
excluding, based on the first difference having a larger magnitude than the second difference, data describing the second predictor variable from the output set of explanatory codes.

6. The non-transitory computer-readable medium of claim 5 , wherein determining the set of maximum predictor values comprising obtaining a respective endpoint of a respective domain of a respective predictor variable as a respective maximum predictor value.

7. The non-transitory computer-readable medium of claim 1 , wherein the set of predictor variables indicate behavioral traits of entities.

8. The non-transitory computer-readable medium of claim 1 , wherein the output risk indicator corresponds to a credit score of the target entity.

9. The non-transitory computer-readable medium of claim 1 , wherein:
the neural network comprises a plurality of nodes organized in connected layers,
a first predictor variable is connected to an output of the neural network via a first set of connections among nodes in the connected layers,
the values of the risk indicator are determined, at least in part, on a first set of operations applied using the first set of connections between the first predictor variable and the output of the neural network,
a second predictor variable is connected to the output of the neural network via a second set of connections among nodes in the connected layers, and
the values of the risk indicator are determined, at least in part, on a second set of operations applied using the second set of connections between the second predictor variable and the output of the neural network.

10. A system comprising:
one or more network interfaces configured for communicating with one or more user devices;
one or more processing devices communicatively coupled to the one or more network interfaces;
one or more memory devices in which instructions executable by the one or more processing devices are stored for causing the one or more processing devices to perform operations comprising:
retrieving an input set of predictor values for a set of predictor variables, wherein the input set of predictor values corresponds to a target entity,
determining an output risk indicator by applying a neural network to the input set of predictor values, wherein the neural network is trained to compute values of a risk indicator from the set of predictor variables, and a monotonic relationship exists between each predictor variable and the risk indicator as computed by the neural network;
determining, based on the neural network, impacts of the set of predictor variables, respectively, on the output risk indicator;
generating, based on the neural network, an output set of explanatory codes based on the determined impacts;
generating one or more electronic communications that include the output risk indicator and the output set of explanatory codes; and
configuring the one or more network interfaces to transmit the one or more electronic communications to the one or more user devices for display, at the one or more user devices, of the output risk indicator and the output set of explanatory codes.

11. The system of claim 10 , wherein determining the impacts of the set of predictor variables on the output risk indicator comprises:
for each predictor variable in the set of predictor variables,
determining a difference between (1) a maximum output value of the neural network generated using an optimized value of the predictor variable and (2) an output value of the neural network generated using an input value of the predictor variable from the input set of predictor va

Assignees

• Equifax Inc

Inventors

• Turner Matthew
• Mcburnett Michael

Classifications

• G06Q40/03

  Credit; Loans; Processing thereof · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/082 Primary

  modifying the architecture, e.g. adding, deleting or silencing nodes or connections · CPC title