Machine-Learning Recurring Charge Detection from Transaction Features

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing technologies for identifying recurring charges on financial accounts are inaccurate, leading to misclassification of transactions and frustrating user experiences due to the need for manual resolution, which can lead to dissatisfaction with financial institutions.

Innovation Solution

A computing platform uses a machine-learning model to analyze transaction data, including transaction-level, merchant-level, and account-merchant-level features, to accurately classify transactions as recurring or non-recurring, and provides interfaces for managing and interacting with identified recurring charges.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional methods are used to identify recurring charges, then the system is simple to operate, but the accuracy of identification is low leading to misclassification

Engineering Contradiction:

Improveaccuracy of identifying recurring chargesVSAvoidcomplexity of classification system

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent introduces a machine learning model as an intermediary between transaction data and classification results. The model processes multiple features (transaction amount, frequency, merchant category, time patterns) and transforms them into accurate recurring charge identification, resolving the contradiction by using a sophisticated intermediary to achieve high accuracy without requiring complex manual analysis systems

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system changes parameters by using multiple feature dimensions (transaction amount, frequency, merchant category, time of day, day of week) rather than single-parameter rules. The machine learning model evaluates combinations of these parameters to dynamically determine recurring charge status, enabling accurate identification while maintaining systematic operation

Inventive Principle:

Principle #35Parameter changes

2Ease of operation

If manual resolution is required for misclassified transactions, then the classification system can be simple, but user experience deteriorates due to frustration and time loss

Engineering Contradiction:

Improveease of transaction managementVSAvoidtime required for manual resolution

Core Design Contradiction:

Ease of operationVSLoss of time

Solution Approach 1:

The system provides self-service capabilities by automatically identifying and classifying recurring charges without requiring user intervention. The machine learning model autonomously processes transactions, and the system presents organized recurring charge information to users who can review and manage their recurring transactions without manual analysis or resolution processes

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system implements feedback mechanisms where classification results are presented to users with sufficient information (merchant name, amount, frequency, next charge date) enabling users to verify and manage their recurring charges. This feedback loop eliminates the need for manual resolution by providing transparent, actionable classification results

Inventive Principle:

Principle #23Feedback

3Measurement precision

If comprehensive transaction analysis is performed to improve accuracy, then identification precision increases, but processing time and computational resources increase

Engineering Contradiction:

Improveaccuracy of transaction classificationVSAvoidprocessing time for transaction analysis

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary actions by pre-processing transaction data into meaningful features (amount, frequency, merchant category, time patterns) before classification. The machine learning model is trained in advance on historical data, enabling it to quickly evaluate new transactions using pre-learned patterns, thus achieving high accuracy without excessive processing time for each individual transaction

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system segments the transaction analysis process into distinct feature extraction components (transaction-level features, account-merchant-level features) that can be processed independently. This segmentation allows parallel processing of different feature types, reducing overall processing time while maintaining comprehensive analysis for accurate classification

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS12361415B2Computing systems and methods for identifying and providing information about recurring transactions

Publication Date: 2025.07.15 CAPITAL ONE FINANCIAL CORP

AI summary

A computing platform is configured with new software technology for accurately identifying recurring charges and facilitating interaction with identified recurring charges, which may involve functionality for (a) obtaining data for a given transaction involving a given customer account and a given merchant, (b) applying pre-processing logic to the obtained data for the given transaction and thereby derive feature data for the given transaction, (c) inputting the feature data for the given transaction into a trained machine-learning model that functions to (i) evaluate the feature data for the given transaction and (ii) based on the evaluation, output a score for the given transaction that indicates a likelihood that the given transaction is a recurring charge, and (d) based on the score for the given transaction, determining whether to classify the given transaction as a recurring charge.