Granted Patent B2

US 12,743,733 · App. 19/008,111 · Granted Sep 22, 2026

Inventors: Rama Krishnam Raju Rudraraju (Hyderabad, IN); Om Purushotham Akarapu (Hyderabad, IN); Adluri Harshitha (Hyderabad, IN); Sathish Kondabathula (Hyderabad, IN)

Assignee: BANK OF AMERICA CORPORATION

G06Q50/265G06Q10/0635G06Q20/40G06Q40/12G06Q40/00G06Q40/02

Abstract

Claims (71)

1 . A system for requirements scoring and malfeasance detection via machine learning, the system comprising:

a processing device; and

a non-transitory storage device containing instructions, when executed by the processing device, the instructions cause the processing device to perform the steps of:

receiving resource transfer data associated with an account identifier, the resource transfer data comprising transfer entries from a plurality of channels, wherein each of the transfer entries comprises a string element describing the transfer entry;

determining an obligation record in the resource transfer data based on the transfer entries comprising an obligation transaction, wherein determining the obligation record comprises referencing a database of predetermined text samples associated with obligation records and parsing a string element of each transfer entry to detect, relative to at least one of the predetermined text samples, at least one of: (i) an equality match, (ii) a substring match, (iii) a regular-expression match, and (iv) a fuzzy match determined using a string-distance metric;

segmenting, for each of the transfer entries, the string element into discrete tokens;

normalizing the discrete tokens into root tokens using lemmatization;

grouping related root tokens into grouped root tokens via fuzzy logic, by assigning, using a membership function, a degree of membership of each root token to one or more fuzzy sets and aggregating the related root tokens into the grouped root tokens using one or more fuzzy operators;

transforming the grouped root tokens into categorical feature vectors, by mapping each grouped root token to an index in a vocabulary and generating a one-hot, multi-hot, or probabilistic vector representation based on the degrees of membership;

determining, using a machine learning model comprising an ensemble learning model, a probability of each of the categorical feature vectors being associated with requirements;

tagging, as requirements resource transfers, the categorical feature vectors upon a first condition where the probability is above a predetermined threshold;

storing the requirements resource transfers in a repository;

determining, using random forest regression model trained to apply a requirements calculation model, an aggregate requirements index for the requirements resource transfers by receiving respective categorical feature vectors;

determining the aggregate requirements index of the requirements resource transfers at a predetermined interval; and

identifying, using the machine learning model, requirements-directed obligation record malfeasance upon a second condition where a change of the aggregate requirements index is above or below a predetermined rate.

2 . The system of claim 1 , wherein the instructions further cause the processing device to perform the steps of:

tagging, as a malfeasance account identifier, the account identifier upon the determining of the requirements-directed obligation record malfeasance;

transmitting the requirements resource transfers associated with the malfeasant account identifier to a second repository for further investigation;

generating an alert signal comprising descriptors for the account identifier and the requirements resource transfers; and

transmitting the alert signal to an endpoint device.

3 . The system of claim 2 , wherein the instructions further cause the processing device to perform the steps of:

receiving, from the endpoint device, a feedback signal generated at the endpoint device in response to accepting or rejecting the identifying of the requirements-directed obligation record malfeasance; and

transmitting the feedback signal to the machine learning model as training data.

4 . The system of claim 3 , wherein the alert signal causes the endpoint device to display an alert banner consisting of at least one of: (i) the account identifier, (ii) the aggregate requirements index, and (iii) a rate of the change of the aggregate requirements index.

5 . The system of claim 4 , wherein the feedback signal is generated as a result of an interaction with the alert banner.

6 . A computer program product for requirements scoring and malfeasance detection via machine learning, the computer program product comprising a non-transitory computer-readable medium comprising code causing an apparatus to:

receive resource transfer data associated with an account identifier, the resource transfer data comprising transfer entries from a plurality of channels, wherein each of the transfer entries comprises a string element describing the transfer entry;

determine an obligation record in the resource transfer data based on the transfer entries comprising an obligation transaction, wherein determining the obligation record comprises referencing a database of predetermined text samples associated with obligation records and parsing a string element of each transfer entry to detect, relative to at least one of the predetermined text samples, at least one of: (i) an equality match, (ii) a substring match, (iii) a regular-expression match, and (iv) a fuzzy match determined using a string-distance metric;

segment, for each of the transfer entries, the string element into discrete tokens;

normalize the discrete tokens into root tokens using lemmatization;

group related root tokens into grouped root tokens via fuzzy logic, by assigning, using a membership function, a degree of membership of each root token to one or more fuzzy sets and aggregating the related root tokens into the grouped root tokens using one or more fuzzy operators;

transform the grouped root tokens into categorical feature vectors, by mapping each grouped root token to an index in a vocabulary and generating a one-hot, multi-hot, or probabilistic vector representation based on the degrees of membership;

determine, using a machine learning model comprising an ensemble learning model, a probability of each of the categorical feature vectors being associated with requirements;

tag, as requirements resource transfers, the categorical feature vectors upon a first condition where the probability is above a predetermined threshold;

store the requirements resource transfers in a repository;

determine, using random forest regression model trained to apply a requirements calculation model, an aggregate requirements index for the requirements resource transfers by receiving respective categorical feature vectors;

determine the aggregate requirements index of the requirements resource transfers at a predetermined interval; and

identify, using the machine learning model, requirements-directed obligation record malfeasance upon a second condition where a change of the aggregate requirements index is above or below a predetermined rate.

7 . The computer program product of claim 6 , wherein the code further causes the apparatus to:

tag, as a malfeasance account identifier, the account identifier upon the determining of the requirements-directed obligation record malfeasance;

transmit the requirements resource transfers associated with the malfeasant account identifier to a second repository for further investigation;

generate an alert signal comprising descriptors for the account identifier and the requirements resource transfers; and

transmit the alert signal to an endpoint device.

8 . The computer program product of claim 7 , wherein the code further causes the apparatus to:

receive, from the endpoint device, a feedback signal generated at the endpoint device in response to accepting or rejecting the identifying of the requirements-directed obligation record malfeasance; and

transmit the feedback signal to the machine learning model as training data.

9 . The computer program product of claim 8 , wherein the alert signal causes the endpoint device to display an alert banner consisting of at least one of: (i) the account identifier, (ii) the aggregate requirements index, and (iii) a rate of the change of the aggregate requirements index.

10 . The computer program product of claim 9 , wherein the feedback signal is generated as a result of an interaction with the alert banner.

11 . A method for requirements scoring and malfeasance detection via machine learning, the method comprising:

receiving resource transfer data associated with an account identifier, the resource transfer data comprising transfer entries from a plurality of channels, wherein each of the transfer entries comprises a string element describing the transfer entry;

determining an obligation record in the resource transfer data based on the transfer entries comprising an obligation transaction, wherein determining the obligation record comprises referencing a database of predetermined text samples associated with obligation records and parsing a string element of each transfer entry to detect, relative to at least one of the predetermined text samples, at least one of: (i) an equality match, (ii) a substring match, (iii) a regular-expression match, and (iv) a fuzzy match determined using a string-distance metric;

segmenting, for each of the transfer entries, the string element into discrete tokens;

normalizing the discrete tokens into root tokens using lemmatization;

grouping related root tokens into grouped root tokens via fuzzy logic, by assigning, using a membership function, a degree of membership of each root token to one or more fuzzy sets and aggregating the related root tokens into the grouped root tokens using one or more fuzzy operators;

transforming the grouped root tokens into categorical feature vectors, by mapping each grouped root token to an index in a vocabulary and generating a one-hot, multi-hot, or probabilistic vector representation based on the degrees of membership;

determining, using a machine learning model comprising an ensemble learning model, a probability of each of the categorical feature vectors being associated with requirements;

tagging, as requirements resource transfers, the categorical feature vectors upon a first condition where the probability is above a predetermined threshold;

storing the requirements resource transfers in a repository;

determining, using random forest regression model trained to apply a requirements calculation model, an aggregate requirements index for the requirements resource transfers by receiving respective categorical feature vectors;

determining the aggregate requirements index of the requirements resource transfers at a predetermined interval; and

identifying, using the machine learning model, requirements-directed obligation record malfeasance upon a second condition where a change of the aggregate requirements index is above or below a predetermined rate.

12 . The method of claim 5 , wherein the method further comprises:

tagging, as a malfeasance account identifier, the account identifier upon the determining of the requirements-directed obligation record malfeasance;

transmitting the requirements resource transfers associated with the malfeasant account identifier to a second repository for further investigation;

generating an alert signal comprising descriptors for the account identifier and the requirements resource transfers; and

transmitting the alert signal to an endpoint device.

13 . The method of claim 12 , wherein the method further comprises:

receiving, from the endpoint device, a feedback signal generated at the endpoint device in response to accepting or rejecting the identifying of the requirements-directed obligation record malfeasance; and

transmitting the feedback signal to the machine learning model as training data.

14 . The method of claim 13 , wherein the alert signal causes the endpoint device to display an alert banner consisting of at least one of: (i) the account identifier, (ii) the aggregate requirements index, (iii) and (iii) a rate of the change of the aggregate requirements index.

15 . The method of claim 14 , wherein the feedback signal is generated as a result of an interaction with the alert banner.

Assignments (1)

Continuity (1)

References Cited (19)