Systems and methods for using a predictive engine to predict failures in machine-learning trained systems

Patent metadata

Field	Value
Publication number	US-2024232915-A9
Application number	US-202318496533-A
Country	US
Kind code	A9
Filing date	Oct 27, 2023
Priority date	Oct 22, 2019
Publication date	Jul 11, 2024
Grant date	—

Abstract

A computer-implemented method for using a machine-learning trained predictive engine to predict failures includes receiving electronic prior transaction data corresponding to a plurality of prior successful transactions and a plurality of prior unsuccessful transactions, and training a machine learning predictive engine based on the plurality of prior successful transactions and the plurality of prior unsuccessful transactions. Electronic transaction data may be received, the electronic transaction data being associated with a user, an item, and candidate transaction terms, the electronic transaction data being associated with a candidate transaction. The machine learning predictive engine may determine a likelihood of success of the candidate transaction based on the electronic transaction data, and display the likelihood of success of the candidate transaction.

First claim

What is claimed is:

1 . A computer-implemented method for improving transaction success, comprising:
receiving electronic transaction data associated with a candidate transaction, the electronic transaction data including (i) information associated with a buyer, a seller, a lender, and an item involved with the candidate transaction, and (ii) candidate terms for the candidate transaction; determining, using a trained machine learning predictive engine, a likelihood of success for the candidate transaction based on the received electronic transaction data, wherein:
the trained machine learning predictive engine has been trained, based on prior electronic transaction data associated with a plurality of prior successful transactions and a plurality of prior unsuccessful transactions as training data, so as to develop a prediction model for likelihood of success for input transactions based on aspects of the input transactions; and
the prior electronic transaction data respectively includes, for each of the prior successful and unsuccessful transactions, information associated with a prior item, a prior buyer, a prior seller, a prior lender, and prior terms; and
causing a user device associated with a user to display the likelihood of success of the candidate transaction.

2 . The computer-implemented method of claim 1 , wherein:
the prediction model for likelihood of success includes respective feature vectors corresponding to each of the plurality of prior successful and unsuccessful transactions; and determining the likelihood of success using the trained machine learning predictive engine includes:
defining the candidate transaction as a further feature vector based on the received electronic transaction data; and
performing a clustering operation for the further feature vector with regard to the predictive model to determine a distance in an N-dimensional space from the further feature vector to a cluster associated with transaction success, the determined distance associated with the likelihood of success for the candidate transaction.

3 . The computer-implemented method of claim 2 , further comprising:
receiving a definition of success for the candidate transaction, wherein the machine learning predictive engine is further configured to define a first cluster associated with transaction failure and a second cluster associated with transaction success based on the received definition of success.

4 . The method of claim 3 , wherein defining the first cluster based on the definition of success for the candidate transaction causes at least one of the prior successful or unsuccessful transactions to be redefined as unsuccessful or successful, respectively, such that the prediction model is tuned to the candidate transaction.

5 . The computer-implemented method of claim 1 , further comprising:
determining whether the likelihood of success is below a predetermined threshold; in response to determining that the likelihood of success is below the predetermined threshold, determining one or more alternate terms to at least a portion of the candidate terms of the candidate transaction that increase the likelihood of success above the predetermined threshold; and causing the user device to display the one or more alternate terms.

6 . The computer-implemented method of claim 1 , further comprising:
determining whether the likelihood of success is below a predetermined threshold; in response to determining that the likelihood of success is below the predetermined threshold, determining one or more alternate items to the item of the candidate transaction that increase the likelihood of success above the predetermined threshold; and causing the user device to display the one or more alternate items.

7 . The computer-implemented method of claim 6 , wherein the one or more alternate items are older than the item.

8 . The computer-implemented method of claim 1 , wherein:
the candidate terms for the candidate transaction include at least one term having a defined range associated with one or more of the buyer, the seller, or the lender; and the computer-implemented method further comprises determining a willingness to haggle associated with the one or more of the buyer, the seller, or the lender based on whether the defined range of the at least one term exceeds a predetermined range.

9 . The computer-implemented method of claim 8 , further comprising:
generating, by the machine learning predictive engine, at least one recommendation with regard to the candidate transaction based, at least in part, on the determined willingness to haggle; and causing the user device to display the at least one recommendation.

10 . The computer-implemented method of claim 1 , further comprising:
comparing the likelihood of success to a first predetermined threshold and a second predetermined threshold lower than the first predetermined threshold; and in response to the comparing, selectively:
in the case of the likelihood of success being between the first predetermined threshold and the second predetermined threshold, determining one or more alternate terms to at least a portion of the candidate terms of the candidate transaction that increase the likelihood of success above the first predetermined threshold, and causing the user device to display the one or more alternate terms; or
in the case of the likelihood of success being below both the first predetermined threshold and the second predetermined threshold, determining one or more alternate items to the item of the candidate transaction that increase the likelihood of success above the second predetermined threshold, and causing the user device to display the one or more alternate items.

11 . A system for improving transaction success, comprising:
at least one memory storing instructions, and a trained machine learning predictive engine that includes a prediction model for likelihood of success for input transactions; and at least one processors operatively connected to the memory, and configured to execute the instructions to perform operations, including:
receiving electronic transaction data associated with a candidate transaction, the electronic transaction data including (i) information associated with a buyer, a seller, a lender, and an item involved with the candidate transaction, and (ii) candidate terms for the candidate transaction;
determining, using the trained machine learning predictive engine, a likelihood of success for the candidate transaction based on the received electronic transaction data, wherein:
the trained machine learning predictive engine has been trained, based on prior electronic transaction data associated with a plurality of prior successful transactions and a plurality of prior unsuccessful transactions as training data, so as to develop the prediction model for likelihood of success for input transactions based on aspects of the input transactions; and
the prior electronic transaction data respectively includes, for each of the prior successful and unsuccessful transactions, information associated with a prior item, a prior buyer, a prior seller, a prior lender, and prior terms; and

causing a user device associated with a user to display the likelihood of success of the candidate transaction.

12 . The system of claim 11 , wherein:
the prediction model for likelihood of success includes respective feature vectors corresponding to each of the plurality of prior successful and unsuccessful transactions; and determining the likelihood of success using the trained machine learning predictive engine includes:
defining the candidate transaction as a further fe

Assignees

• Capital One Services Llc

Inventors

• Dagley Geoffrey
• Anderson Stephen
• Wylie Stephen
• Tang Qiaochu
• Price Micah
• Hoover Jason
• Przano Kristen

Classifications

• G06F18/24147

  Distances to closest patterns, e.g. nearest neighbour classification · CPC title

• G06F18/2137

  based on criteria of topology preservation, e.g. multidimensional scaling or self-organising maps · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06F18/23

  Clustering techniques · CPC title

• G06Q20/405

  Establishing or using transaction specific rules · CPC title