Techniques to predict and implement an amortized bill payment system

Patent metadata

Field	Value
Publication number	US-12260424-B2
Application number	US-202318218949-A
Country	US
Kind code	B2
Filing date	Jul 6, 2023
Priority date	Oct 18, 2019
Publication date	Mar 25, 2025
Grant date	Mar 25, 2025

Abstract

Various embodiments are generally directed to techniques utilizing computers to process service billing data, generate one or more groups of customers based on one or more attributes of the data, and train a recurrent neural network (RNN) to predict future bill amounts for customers. Embodiments further include techniques to predict the future bill amounts of the customers and providing an indication of the bill amounts to the customers.

First claim

What is claimed is:

1. An apparatus, comprising:
a communication interface;
a processor circuit coupled with the communication interface; and
a memory coupled with the processor circuit and the communication interface, the memory storing instructions which when executed by the processor circuit, cause the processor circuit to:
receive, via the communication interface coupled with the processor circuit, service billing data for a plurality of customers, the service billing data comprising a plurality of attributes for each of the plurality of customers;
identify a group of customers based on a cluster operation using at least two attributes of the plurality of attributes;
predict, with a plurality of recurrent neural networks (RNNs), predicted periodic bill amounts for a future period of time for the group of customers, the prediction based on bill information for the group of customers received during a predetermined duration, wherein each of the RNNs is associated with a particular customer in the group of customers and trained with particular bill information for the particular customer to predict periodic bill amounts for the future period of time for the particular customer;
retrain each of the plurality of RNNs with the predicted periodic bill amounts for the future period of time from all of the plurality of customers;
predict, for the particular customer with an associated RNN, the periodic bill amounts for the future period of time, wherein the associated RNN is trained with the bill information for the particular customer and retrained with the predicted periodic bill amounts for the plurality of customers; and
display the periodic bill amounts for the future period of time for the particular customer.

2. The apparatus of claim 1 , wherein the bill information comprises amounts of bills for a specific service over the predetermined duration for each of the plurality of customers, and wherein each of the amounts of bills is a periodic bill amount for a specific service.

3. The apparatus of claim 1 , wherein the plurality of attributes comprises service type information, which comprises a service type identifier to identify a service type of a specific service.

4. The apparatus of claim 1 , wherein the plurality of attributes comprises location information, which comprises location identifiers to identify locations of the plurality of customers, each customer of the plurality of customers associated with a location identifier to identify a location.

5. The apparatus of claim 1 , wherein the cluster operation comprises applying a k-nearest neighbors (k-NN) algorithm on the service billing data to generate two or more groups of customers.

6. The apparatus of claim 5 , wherein the k-NN algorithm to determine the customer of the plurality of customers to include in the group of customers of the two or more groups of customers based on voting of other customers of the group of customers.

7. The apparatus of claim 6 , wherein the voting of the group of customers is based on location information and service type information associated with each of the other customers of the group of customers.

8. The apparatus of claim 1 , wherein the instructions which when executed by the processor circuit, further cause the processor circuit to determine a margin of error amount for an amortized bill amount for the particular customer and adjust the amortized bill amount by the margin of error amount prior to the display.

9. The apparatus of claim 1 , wherein the instructions which when executed by the processor circuit, further cause the processor circuit to:
determine an actual bill amount for a specific service provided for the future period of time for the particular customer;
determine a difference amount between the actual bill amount and a total amortized bill amount for the future period of time for the particular customer, wherein the total amortized bill amount is the total of the periodic bill amounts predicted for the particular customer for the future period of time; and
bill or refund the customer the difference amount based on whether the actual bill amount is greater than or less than the total amortized bill amount.

10. A method comprising:
receiving, by a computer server, service billing data for a plurality of customers, the service billing data comprising a plurality of attributes for each of the plurality of customers;
identifying, by the computer server, a group of customers based on a cluster operation using at least two attributes of the plurality of attributes;
predicting, by the computer server, with a plurality of recurrent neural networks (RNNs), predicted periodic bill amounts for a future period of time for the group of customers, the prediction based on bill information for the group of customers received during a predetermined duration, wherein each of the RNNs is associated with a particular customer in the group of customers and trained with particular bill information for the particular customer to predict periodic bill amounts for the future period of time for the particular customer;
retraining, by the computer server, each of the plurality of RNNs with the predicted periodic bill amounts for the future period of time from all of the plurality of customers;
predicting, by the computer server, for the particular customer with an associated RNN, the periodic bill amounts for the future period of time, wherein the associated RNN is trained with the bill information for the particular customer and retrained with the predicted periodic bill amounts for the plurality of customers; and
causing, by the computer server, the periodic bill amounts to be displayed for the future period of time for the particular customer.

11. The method of claim 10 , wherein the bill information comprises amounts of bills for a specific service over the predetermined duration for each of the plurality of customers, and wherein each of the amounts of bills is a periodic bill amount for a specific service.

12. The method of claim 10 , wherein the plurality of attributes comprises service type information, which comprises a service type identifier to identify a service type of a specific service.

13. The method of claim 10 , wherein the plurality of attributes comprises location information, which comprises location identifiers to identify locations of the plurality of customers, each customer of the plurality of customers associated with a location identifier to identify a location.

14. The method of claim 10 , wherein the cluster operation comprises applying a k-nearest neighbors (k-NN) algorithm on the service billing data to generate two or more groups of customers.

15. The method of claim 14 , wherein the k-NN algorithm to determine the customer of the plurality of customers to include in the group of customers of the two or more groups of customers based on voting of other customers of the group of customers.

16. The method of claim 15 , wherein the voting of the group of customers is based on location information and service type information associated with each of the other customers of the group of customers.

17. A non-transitory computer-readable storage medium, the computer-readable storage medium including instructions that when executed by a computer, cause the computer to:
receive service billing data for a plurality of customers, the service billing data comprising a plurality of attributes for each of the plurality of customers;
identify a group of customers based on a cluster operation using at least two attributes of the plurality of attributes;
predict

Assignees

• Capital One Services Llc

Inventors

• Walters Austin Grant
• Pham Vincent
• Farivar Reza
• Goodsitt Jeremy Edward

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06Q20/102

  Bill distribution or payments · CPC title

• G06N3/08

  Learning methods · CPC title

• G06Q20/0855

  involving a third party · CPC title