Optimizing neural networks for generating analytical or predictive outputs

Patent metadata

Field	Value
Publication number	US-2018025273-A1
Application number	US-201715724828-A
Country	US
Kind code	A1
Filing date	Oct 4, 2017
Priority date	Mar 27, 2015
Publication date	Jan 25, 2018
Grant date	—

Abstract

Certain embodiments involve generating or optimizing a neural network for generating analytical or predictive outputs. 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 response variable. 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 response variable. The optimized neural network can be used both for accurately determining response variables 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 response variable. The neural network can be used to generate adverse action codes upon which consumer behavior can be modified to improve the response variable score.

First claim

What is claimed is:

1 . A system comprising:
a processing device; and a memory device in which instructions executable by the processing device are stored for causing the processing device to:
receive a plurality of predictor variables, wherein each predictor variable corresponds to an entity;
determine a correlation between each predictor variable and an outcome;
generate a neural network that includes a hidden layer for determining a relationship between each predictor variable and a response variable based on the correlation, wherein the response variable indicates a behavior associated with the entity and wherein the neural network is operable for determining whether a monotonic relationship exists between each predictor variable and the response variable; and
iteratively adjust the neural network so that the monotonic relationship exists between each predictor variable and the response variable as determined by the neural network.

2 . The system of claim 1 , wherein the processing device is configured to:
adjust the neural network by adjusting at least one of a number of nodes in the hidden layer of the neural network, a predictor variable in the plurality of predictor variables, or a number of layers in the hidden neural network; determine, using the neural network, the response variable based at least partially on the predictor variables after the monotonic relationship exists between each predictor variable and the response variable; determine, based on a rate of change of the response variable with respect to the predictor variable, that the monotonic relationship exists between each predictor variable and the response variable; determine, using the neural network, an impact of each predictor variable on the response variable; and generate, using the neural network, an adverse action code associated with each predictor variable that indicates the impact of each predictor variable on the response variable.

3 . The system of claim 2 , wherein the hidden layer comprises at least two hidden layers.

4 . The system of claim 1 , wherein the processing device is configured to determine the correlation between each predictor variable and an outcome by determining a correlation between each predictor variable and an amount of positive outcomes or negative outcomes, wherein each positive outcome indicates that a condition is satisfied and each negative outcome indicates failure to satisfy the condition

5 . The system of claim 4 , wherein the processing device is configured to determine the correlation between each predictor variable and the amount of positive outcomes or negative outcomes by performing operations comprising verifying that a bivariate relationship exists between each predictor variable and the amount of positive or negative outcomes.

6 . The system of claim 2 , wherein the processing device is further configured to determine a rank of each predictor variable, using the neural network, based on the impact of each predictor variable on the response variable.

7 . The system of claim 1 , wherein the response variable corresponds to a credit score of the entity.

8 . A method comprising:
receiving, by a processor, a plurality of predictor variables, wherein each predictor variable corresponds to an entity; determining, by the processor, a correlation between each predictor variable and an amount of positive outcomes or negative outcomes, wherein each positive outcome indicates that a condition is satisfied and each negative outcome indicates failure to satisfy the condition; generating, by the processor, a neural network that includes a hidden layer for determining a relationship between each predictor variable and a response variable based on the correlation, wherein the response variable is indicates a behavior associated with the entity; and iteratively adjusting the neural network so that a monotonic relationship exists between each predictor variable and the response variable as determined by the neural network.

9 . The method of claim 8 , wherein iteratively adjusting the neural network includes determining whether the monotonic relationship exists between each predictor variable and the response variable.

10 . The method of claim 8 , wherein adjusting the neural network includes adjusting at least one of a number of nodes in the hidden layer of the neural network, a predictor variable in the plurality of predictor variables, or a number of layers in the hidden neural network, wherein the method further comprises:
determining, using the neural network, the response variable based at least partially on the predictor variables after the monotonic relationship exists between each predictor variable and the response variable; determining, based on a rate of change of the response variable with respect to the predictor variable, that the monotonic relationship exists between each predictor variable and the response variable; determining, using the neural network, an impact of each predictor variable on the response variable; determining, using the neural network, an adverse action code associated with each predictor variable that indicates the impact of each predictor variable on the response variable; and determining, using the neural network, a rank of each predictor variable based on the impact of each predictor variable on the response variable.

11 . The method of claim 10 , further comprising outputting, by the processor, the response variable, each predictor variable, the adverse action code associated with each predictor variable, and the rank of each predictor variable.

12 . The method of claim 10 , wherein the hidden layer comprises at least two hidden layers.

13 . The method of claim 8 , wherein determining the correlation between each predictor variable and the amount of positive outcomes or negative outcomes includes verifying that a bivariate relationship exists between each predictor variable and the amount of positive or negative outcomes.

14 . A non-transitory computer-readable storage medium having program code that is executable by a processor device to cause a computing device to perform operation, the operations comprising:
receiving a plurality of predictor variables, wherein each predictor variable corresponds to an entity; determining a correlation between each predictor variable and an amount of positive outcomes or negative outcomes, wherein each positive outcome indicates that a condition is satisfied and each negative outcome indicates failure to satisfy the condition; generating a neural network that includes a hidden layer for determining a relationship between each predictor variable and a response variable based on the correlation, wherein the response variable indicates a behavior associated with the entity and wherein the neural network is operable for determining whether a monotonic relationship exists between each predictor variable and the response variable; and iteratively adjusting the neural network so that the monotonic relationship exists between each predictor variable and the response variable as determined by the neural network.

15 . The non-transitory computer-readable medium of claim 14 , wherein the operations further comprise:
adjusting the neural network by adjusting at least one of a number of nodes in the hidden layer of the neural network, a predictor variable in the plurality of predictor variables, or a number of layers in the hidden neural network; determining, using the neural network, the response variable based at least partially on the predictor variables after the monotonic relationship exists b

Assignees

• Equifax Inc

Inventors

• Jordan Lewis
• 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