# US 12,002,091 B2 · Technology system for assisting financial institutions in debt collection

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 12,002,091 B2
- Document ID: US12002091B2
- Status: Granted
- Kind: B2
- Application number: 17/659,017
- Filed: 2022-04-13
- Published: 2023-06-08
- Granted: 2024-06-04
- Earliest claimed: 2021-12-03
- Assignees: Oracle Financial Services Software Limited
- Applicants: Oracle Financial Services Software Limited
- Inventors: Rahul Yadav; Veresh Jain; Shubham Negi; Rahul Kiran Gaddam; Phanindra Choda
- Web page: https://patentgrep.com/patent/US12002091B2

## CPC

- G06Q 40/03: Credit; Loans; Processing thereof
- G06F 18/214: Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F 18/23: Clustering techniques
- G06N 3/092: Reinforcement learning
- G06N 20/00: Machine learning
- G06Q 40/02: Banking, e.g. interest calculation or account maintenance (credit or loans G06Q40/03)
- G06N 5/01: Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- G06N 20/10: using kernel methods, e.g. support vector machines [SVM]
- G06N 20/20: Ensemble learning
- G06Q: INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

According to an aspect, a technology system maintains an account data specifying the details of multiple accounts including a specific set of accounts as being delinquent. Upon receiving a feature set to be used, the system clusters the multiple accounts into groups based on the feature set. The system generates a multi-classification model for assignment of agents to the accounts based on the group qualifiers associated with the groups and the feature set. The system then identifies a target group likely to present problems with debt collection, assigns, using the multi-classification model, a corresponding agent for each account of the group and determines a recommended recovery strategy for each assignment. The system provides the details of the group, the corresponding agents assigned and the recommended recovery strategy for each assignment.

## Claims

### Claim 1

A non-transitory machine-readable medium storing one or more sequences of instructions for assisting financial institutions in debt collection, wherein execution of said one or more instructions by one or more processors contained in a digital processing system cause said digital processing system to perform the actions of:
maintaining an account data specifying account attributes of a plurality of accounts, said account data indicating a specific set of accounts of said plurality of accounts as being delinquent;
receiving a feature set comprising a set of account attributes of said account attributes;
clustering, using a first machine learning (ML) approach, said plurality of accounts to determine a set of groups, wherein said first ML approach operates to group accounts based on said account attributes while giving higher weightage to said set of account attributes, each group of said set of groups being associated with a respective set of group qualifiers, each group qualifier specifying a corresponding range of values for a respective attribute of said account attributes,
wherein a first clustering ML approach is used as said first ML approach if a desired count of a number of groups in said set of groups is received with said feature set, and a second clustering ML approach is used if said desired count is not received;
generating, using a second ML approach, a multi-classification model for assignment of a plurality of agents to respective accounts, wherein said second ML approach correlates a set of recovery attributes of each agent and said respective set of group qualifiers, while giving higher weightage to said set of account attributes contained in said feature set;
identifying a target group of said set of groups containing a material number of said specific set of accounts indicated to be delinquent and a subset of accounts which are indicated not to be delinquent in said account data;
assigning, using said multi-classification model, a corresponding agent for each account in said subset of accounts;
determining a recommended recovery strategy for each assignment of corresponding agent and account contained in said subset of accounts; and
providing details of said subset of accounts, said corresponding agent assigned to each of said subset of accounts and said recommended recovery strategy for each assignment.

### Claim 2 (depends on 1)

The non-transitory machine-readable medium of claim 1,
wherein said second clustering ML approach is used if said desired count is not received and if said account data does not contain nested attributes,
wherein a third clustering ML approach is used if said desired count is not received and if said account data contains nested attributes,
wherein said clustering comprises:
generating using said first ML approach, a ML clustering model based on said account data and said feature set; and
predicting said set of groups based on said ML clustering model.

### Claim 3 (depends on 2)

The non-transitory machine-readable medium of claim 2, wherein said first clustering ML approach is K-Means technique,
wherein said second clustering ML approach is O-Clustering technique and
wherein said third clustering ML approach is Expectation maximization technique.

### Claim 4 (depends on 1)

The non-transitory machine-readable medium of claim 1, wherein said account data also specifies a corresponding delinquency status for each account,
wherein said identifying identifies a probability of delinquency for said first group based on weights associated with the corresponding delinquency statuses of the accounts contained in said target group.

### Claim 5 (depends on 1)

The non-transitory machine-readable medium of claim 1, wherein said generating comprises one or more instructions for:
training a plurality of machine learning (ML) classification models based on said set of recovery attributes of each agent of said plurality of agents and said respective set of group qualifiers of said set of groups; and
selecting a suitable ML classification model from said plurality of ML classification models as said multi-classification model.

### Claim 6 (depends on 5)

The non-transitory machine-readable medium of claim 5, wherein said plurality of ML classification models employs one or more of logistic regression, SVMs, ensemble trees, boosting trees, naive Bayes, KNN (k-nearest neighbors) and shallow neural networks,
wherein said selecting said suitable ML classification model is performed based on one or more of confusion matrix, accuracy, precision, recall, F1 score, ROC Curve, and gain and lift chart.

### Claim 7 (depends on 1)

The non-transitory machine-readable medium of claim 1, wherein said determining comprises:
maintaining a reinforcement learning (RL) state table correlating a set of states to a set of strategies, said set of states representing corresponding combinations of group qualifiers and agent attributes; and
identifying, for a first assignment of a first agent and a first account contained in said first group, a first strategy of said set of strategies as said recommended recovery strategy using said RL state table, a group qualifier associated with said first group and an agent attribute associated with said first agent,
wherein the actions further comprise:
displaying on a display unit, said set of groups organized in the form of a hierarchy of nodes with each node corresponding to a group such that said hierarchy represents output of said first ML approach,
wherein each target group in said set of groups is represented as a corresponding leaf node of said hierarchy.

### Claim 8 (depends on 7)

The non-transitory machine-readable medium of claim 7, further comprising one or more instructions for:
receiving from said first agent, a feedback on the success or failure of said first strategy with said first account; and
updating said RL state table to reflect said feedback.

### Claim 9 (depends on 8)

The non-transitory machine-readable medium of claim 8, wherein said RL state table is a Q-table employing Q-Learning algorithm.

### Claim 10

A computer implemented method for assisting financial institutions in debt collection, said method being performed in a digital processing system, said method comprising:
maintaining in a memory of said digital processing system, an account data specifying account attributes of a plurality of accounts, said account data indicating a specific set of accounts of said plurality of accounts as being delinquent;
receiving via a user interface module, a feature set comprising a set of account attributes of said account attributes;
providing in said digital processing system a first machine learning (ML) approach and a second ML approach by execution of corresponding instructions;
clustering, by an account segmentation module using said first ML approach, said plurality of accounts to determine a set of groups, wherein said first ML approach operates to group accounts based on said account attributes while giving higher weightage to said set of account attributes, each group of said set of groups being associated with a respective set of group qualifiers, each group qualifier specifying a corresponding range of values for a respective attribute of said account attributes;
generating, by an agent assignment module using said second ML approach, a multi-classification model for assignment of a plurality of agents to respective accounts, wherein said second ML approach correlates a set of recovery attributes of each agent and said respective set of group qualifiers, while giving higher weightage to said set of account attributes contained in said feature set;
identifying by said account segmentation module, a target group of said set of groups likely to present problems with debt collection, wherein said target group contains a material number of said specific set of accounts indicated to be delinquent and a subset of accounts which are indicated not to be delinquent in said account data;
assigning, by said agent assignment module using said multi-classification model, a corresponding agent for each account in said subset of accounts;
determining, by a strategy recommendation module, a recommended recovery strategy for each assignment of corresponding agent and account contained in said subset of accounts; and
providing details of said subset of accounts, said corresponding agent assigned to each of said subset of accounts and said recommended recovery strategy for each assignment.

### Claim 11 (depends on 10)

The method of claim 10, wherein said clustering comprises:
generating a machine learning (ML) clustering model based on said account data and said feature set; and
predicting said set of groups based on said ML clustering model.

### Claim 12 (depends on 10)

The method of claim 10, wherein said account data also specifies a corresponding delinquency status for each account,
wherein said identifying identifies a probability of delinquency for said first group based on weights associated with the corresponding delinquency statuses of the accounts contained in said target group.

### Claim 13 (depends on 10)

The method of claim 10, wherein said generating comprises:
training a plurality of machine learning (ML) classification models based on said set of recovery attributes of each agent of said plurality of agents and said respective set of group qualifiers of said set of groups; and
selecting a suitable ML classification model from said plurality of ML classification models as said multi-classification model.

### Claim 14 (depends on 10)

The method of claim 10, wherein said determining comprises:
maintaining a reinforcement learning (RL) state table correlating a set of states to a set of strategies, said set of states representing corresponding combinations of group qualifiers and agent attributes; and
identifying, for a first assignment of a first agent and a first account contained in said first group, a first strategy of said set of strategies as said recommended recovery strategy using said RL state table, a group qualifier associated with said first group and an agent attribute associated with said first agent.

### Claim 15 (depends on 14)

The method of claim 14, further comprising:
receiving from said first agent, a feedback on the success or failure of said first strategy with said first account; and
updating said RL state table to reflect said feedback.

### Claim 16

A digital processing system comprising:
a random access memory (RAM) to store instructions for assisting financial institutions in debt collection; and
one or more processors to retrieve and execute the instructions, wherein execution of the instructions causes the digital processing system to perform the actions of:
maintaining an account data specifying account attributes of a plurality of accounts, said account data indicating a specific set of accounts of said plurality of accounts as being delinquent;
receiving a feature set comprising a set of account attributes of said account attributes;
clustering, using a first machine learning (ML) approach, said plurality of accounts to determine a set of groups, wherein said first ML approach operates to group accounts based on said account attributes while giving higher weightage to said set of account attributes, each group of said set of groups being associated with a respective set of group qualifiers, each group qualifier specifying a corresponding range of values for a respective attribute of said account attributes,
wherein a first clustering ML approach is used as said first ML approach if a desired count of a number of groups in said set of groups is received with said feature set, and a second clustering ML approach is used if said desired count is not received;
generating, using a second ML approach, a multi-classification model for assignment of a plurality of agents to respective accounts, wherein said second ML approach correlates a set of recovery attributes of each agent and said respective set of group qualifiers, while giving higher weightage to said set of account attributes contained in said feature set;
identifying a target group of said set of groups containing a material number of said specific set of accounts indicated to be delinquent and a subset of accounts which are indicated not to be delinquent in said account data;
assigning, using said multi-classification model, a corresponding agent for each account in said subset of accounts;
determining a recommended recovery strategy for each assignment of corresponding agent and account contained in said subset of accounts; and
providing details of said subset of accounts, said corresponding agent assigned to each of said subset of accounts and said recommended recovery strategy for each assignment.

### Claim 17 (depends on 16)

The digital processing system of claim 16,
wherein said second clustering ML approach is used if said desired count is not received and if said account data does not contain nested attributes,
wherein a third clustering ML approach is used if said desired count is not received and if said account data contains nested attributes,
wherein for said clustering, said digital processing system performs the actions of:
generating using said first ML approach, a ML clustering model based on said account data and said feature set; and
predicting said set of groups based on said ML clustering model.

### Claim 18 (depends on 16)

The digital processing system of claim 16, wherein said account data also specifies a corresponding delinquency status for each account,
wherein said identifying identifies a probability of delinquency for said first group based on weights associated with the corresponding delinquency statuses of the accounts contained in said target group.

### Claim 19 (depends on 16)

The digital processing system of claim 16, wherein for said generating, said digital processing system performs the actions of:
training a plurality of machine learning (ML) classification models based on said set of recovery attributes of each agent of said plurality of agents and said respective set of group qualifiers of said set of groups; and
selecting a suitable ML classification model from said plurality of ML classification models as said multi-classification model.

### Claim 20 (depends on 16)

The digital processing system of claim 16, wherein said for determining, said digital processing system performs the actions of:
maintaining a reinforcement learning (RL) state table correlating a set of states to a set of strategies, said set of states representing corresponding combinations of group qualifiers and agent attributes; and
identifying, for a first assignment of a first agent and a first account contained in said first group, a first strategy of said set of strategies as said recommended recovery strategy using said RL state table, a group qualifier associated with said first group and an agent attribute associated with said first agent.