Granted Patent B2

US 12,406,213 · App. 18/092,500 · Granted Sep 2, 2025

Inventors: Saman Baghestani (Plano, TX); Rohan Shah (Frisco, TX); Nicholas E. Dolle (Mckinney, TX)

Assignee: Capital One Services, LLC

G06Q10/04G06F18/232G06F18/23213G06N20/00G06Q30/0201G06Q30/0206

Abstract

Claims (53)

1. A computer implemented method for increasing accuracy of weighing a plurality of attributes in a decision model, and revealing which dimension is most impactful in the decision model, the computer implemented method comprising:

concurrently executing, by a plurality of hypervisors, multiple instances of a plurality of operating systems to provide hardware virtualization;

performing block virtualization, by a virtualized storage component of the hardware virtualization, that separates logical storage from physical storage to allow attributes of a data store to be accessed without regard to a heterogeneous structure;

receiving, at a profile engine, the attributes accessed from the data store;

performing file virtualization, using a virtualized storage component of the hardware virtualization, to eliminate dependencies between the accessed attributes at a file-level and a location where files are physically stored;

generating, by the profile engine, a multi-dimensional graph of a plurality of n-dimensions where each dimension of the n-dimensions corresponds with a respective attribute of the attributes;

generating, by the profile engine, a bell curve along the n-dimensions of the multi-dimensional graph;

determining, by the profile engine, which ones of the attributes impact the decision model more than a threshold amount according to a comparison of the attributes against the generated bell curve at a specified value for each of the attributes;

identifying, by unsupervised machine learning, clusters of variable K to each of the attributes processed by K-means clustering that assigns each of the clusters to each of a plurality of K groups based on feature weights;

positioning, by the profile engine on the multi-dimensional graph, the identified clusters, along a first axis comprising the n-dimensions and, along a second axis representing the feature weights, based on how much weight to give to each of the attributes along each of the n-dimensions;

computing, by the profile engine, a distribution of each graph of the multi-dimensional graph along the first and second axes indicating how statistically reliable each of the n-dimensions are in determining predictable patterns of characterization for each of the identified clusters in response to respective combinations of the n-dimensions;

recalibrating, by the profile engine, the feature weights of the attributes for the decision model, as the profile engine receives more data for the attributes according to a continuous decrease in variance between observed and expected outcomes of the decision model to improve accuracy over time of the weighing of each of the attributes within each of the n-dimensions; and

generating, by the profile engine, along dimension lines of the multi-dimensional graph, a plurality of maximum nodes connected to a plurality of edges of a maximum value for each dimension of the n-dimensions along the dimension lines, indicating which particular dimension of the n-dimensions is most impactful in the decision model, according to said particular dimension having a higher value of its node up to the maximum nodes.

2. The computer implemented method of claim 1 , wherein the decision model corresponds to generating a financing structure.

3. The computer implemented method of claim 1 , where the attributes include: a final sale price of a specified product, a breakdown of a sale price, an amount of down payment being provided, condition of the specified product, whether the specified product is new or used, and a type of the specified product.

4. The computer implemented method of claim 1 , wherein the n-dimensions include: a percentage of backend value as compared to a total price of a specified product, frequency of counteroffers presented by sellers, weather on a day of a sale, an age of the specified product, and a color of the specific product.

5. The computer implemented method of claim 1 , wherein the profile engine is a dealer profile engine.

6. A system for increasing accuracy of weighing a plurality of attributes in a decision model, and revealing which dimension is most impactful in the decision model, the system comprising:

a memory storing instructions;

a processor in communication with the memory to execute the instructions to allow:

concurrently executing, by a plurality of hypervisors, multiple instances of a plurality of operating systems to provide hardware virtualization;

performing block virtualization, by a virtualized storage component of the hardware virtualization, that separates logical storage from physical storage to allow attributes of a data store to be accessed without regard to a heterogeneous structure;

receiving, at a profile engine, the attributes accessed from the data store;

performing file virtualization, using a virtualized storage component of the hardware virtualization, to eliminate dependencies between the accessed attributes at a file-level and a location where files are physically stored;

generating, by the profile engine, a multi-dimensional graph of a plurality of n-dimensions where each dimension of the n-dimensions corresponds with a respective attribute of the attributes;

generating, by the profile engine, a bell curve along the n-dimensions of the multi-dimensional graph;

determining, by the profile engine, which ones of the attributes impact the decision model more than a threshold amount according to a comparison of the attributes against the generated bell curve at a specified value for each of the attributes;

identifying, by unsupervised machine learning, clusters of variable K to each of the attributes processed by K-means clustering that assigns each of the clusters to each of a plurality of K groups based on feature weights;

positioning, by the profile engine on the multi-dimensional graph, the identified clusters, along a first axis comprising the n-dimensions and, along a second axis representing the feature weights, based on how much weight to give to each of the attributes along each of the n-dimensions;

computing, by the profile engine, a distribution of each graph of the multi-dimensional graph along the first and second axes indicating how statistically reliable each of the n-dimensions are in determining predictable patterns of characterization for each of the identified clusters in response to respective combinations of the n-dimensions;

recalibrating, by the profile engine, the feature weights of the attributes for the decision model, as the profile engine receives more data for the attributes according to a continuous decrease in variance between observed and expected outcomes of the decision model to improve accuracy over time of the weighing of each of the attributes within each of the n-dimensions; and

generating, by the profile engine, along dimension lines of the multi-dimensional graph, a plurality of maximum nodes connected to a plurality of edges of a maximum value for each dimension of the n-dimensions along the dimension lines, indicating which particular dimension of the n-dimensions is most impactful in the decision model, according to said particular dimension having a higher value of its node up to the maximum nodes.

7. The system of claim 6 , wherein the decision model corresponds to generating a financing structure.

8. The system of claim 6 , where the attributes include: a final sale price of a specified product, a breakdown of a sale price, an amount of down payment being provided, condition of the specified product, whether the specified product is new or used, and a type of the specified product.

9. The system of claim 6 , wherein the n-dimensions include: a percentage of backend value as compared to a total price of a specified product, frequency of counteroffers presented by sellers, weather on a day of a sale, an age of the specified product, and a color of the specific product.

10. The system of claim 6 , wherein the profile engine is a dealer profile engine.

11. A non-transitory computer-readable medium storing instructions that when executed by one or more processors allow:

concurrently executing, by a plurality of hypervisors, multiple instances of a plurality of operating systems to provide hardware virtualization;

performing block virtualization, by a virtualized storage component of the hardware virtualization, that separates logical storage from physical storage to allow attributes of a data store to be accessed without regard to a heterogeneous structure;

receiving, at a profile engine, the attributes accessed from the data store;

performing file virtualization, using a virtualized storage component of the hardware virtualization, to eliminate dependencies between the accessed attributes at a file-level and a location where files are physically stored;

generating, by the profile engine, a multi-dimensional graph of a plurality of n-dimensions where each dimension of the n-dimensions corresponds with a respective attribute of the attributes;

generating, by the profile engine, a bell curve along the n-dimensions of the multi-dimensional graph;

determining, by the profile engine, which ones of the attributes impact a decision model more than a threshold amount according to a comparison of the attributes against the generated bell curve at a specified value for each of the attributes;

identifying, by unsupervised machine learning, clusters of variable K to each of the attributes processed by K-means clustering that assigns each of the clusters to each of a plurality of K groups based on feature weights;

positioning, by the profile engine on the multi-dimensional graph, the identified clusters, along a first axis comprising the n-dimensions and, along a second axis representing the feature weights, based on how much weight to give to each of the attributes along each of the n-dimensions;

computing, by the profile engine, a distribution of each graph of the multi-dimensional graph along the first and second axes indicating how statistically reliable each of the n-dimensions are in determining predictable patterns of characterization for each of the identified clusters in response to respective combinations of the n-dimensions;

recalibrating, by the profile engine, the feature weights of the attributes for the decision model, as the profile engine receives more data for the attributes according to a continuous decrease in variance between observed and expected outcomes of the decision model to improve accuracy over time of weighing of each of the attributes within each of the n-dimensions; and

generating, by the profile engine, along dimension lines of the multi-dimensional graph, a plurality of maximum nodes connected to a plurality of edges of a maximum value for each dimension of the n-dimensions along the dimension lines, indicating which particular dimension of the n-dimensions is most impactful in the decision model, according to said particular dimension having a higher value of its node up to the maximum nodes.

12. The non-transitory computer-readable medium of claim 11 , wherein the decision model corresponds to generating a financing structure.

13. The non-transitory computer-readable medium of claim 11 , where the attributes include: a final sale price of a specified product, a breakdown of a sale price, an amount of down payment being provided, condition of the specified product, whether the specified product is new or used, and a type of the specified product.

14. The non-transitory computer-readable medium of claim 11 , wherein the n-dimensions include: a percentage of backend value as compared to a total price of a specified product, frequency of counteroffers presented by sellers, weather on a day of a sale, an age of the specified product, and a color of the specific product.

15. The non-transitory computer-readable medium of claim 11 , wherein the profile engine is a dealer profile engine.

Assignments (1)

Continuity (2)

References Cited (67)