Collusive Transaction Fraud Detection Using Deep Learning Baselines
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting transaction fraud focus on single entities, failing to detect collusive transaction fraud involving multiple entities, such as money laundering, interchange abuse, and illicit activities.
Innovation Solution
A computer-implemented method and system using deep learning models to generate merchant baselines based on transaction and time series data, identifying related entities, and classifying them into risk groups to detect collusive fraud by comparing transaction and time series data with baselines, and initiating investigation protocols.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing fraud detection methods focus on single entities, then detection simplicity is maintained, but collusive transaction fraud involving multiple entities cannot be detected
Solution Approach 1:
The patent merges multiple entity assessments into a unified fraud detection framework. It combines transaction data, time series data, and entity relationship data from multiple merchants and acquirers into a single analytical model that generates group risk scores, enabling detection of collusive fraud while maintaining systematic coherence.
Solution Approach 2:
The detection system is designed to handle both single-entity and multi-entity fraud scenarios universally. The same baseline generation and scoring mechanisms apply to individual merchants and acquirers, while also functioning collectively to identify collusive relationships, making the system adaptable to various fraud patterns without requiring separate specialized systems.
2Reliability
If deep learning models compare transaction and time series data with baselines for multiple entities, then collusive fraud detection capability is improved, but computational complexity increases
Solution Approach 1:
The computational process is segmented into distinct phases: baseline generation from historical data, real-time data extraction and comparison, risk score calculation, and entity relationship analysis. This segmentation allows each computational task to be optimized independently and executed in a manageable sequence, reducing overall computational burden while maintaining detection accuracy.
Solution Approach 2:
Baselines for transaction data and time series data are generated in advance from historical information before actual fraud detection is needed. This preliminary action pre-computes expected patterns and thresholds, so that during real-time operation, the system only needs to compare current data against these pre-established baselines, significantly reducing computational complexity during critical detection moments.
3Object-affected harmful factors
If related entities are identified and classified into risk groups, then fraud prevention effectiveness is improved, but system operation complexity increases
Solution Approach 1:
The system implements feedback loops where risk scores and classifications are continuously updated based on new transaction data and entity relationships. When collusive fraud is detected, the system provides feedback by placing holds on transactions and generating alerts, which can trigger further investigation or corrective actions, creating a closed-loop system that adapts and improves over time.
Solution Approach 2:
The system automatically performs entity identification, risk scoring, and classification without requiring manual intervention. The deep learning models autonomously analyze data patterns, identify related entities, and assign risk group classifications, reducing operational complexity despite the sophisticated analysis performed.
4Measurement precision
If merchant category codes are modified to correct miscodings, then data accuracy is improved, but processing time increases
Solution Approach 1:
Merchant category code validation and correction are performed as preliminary steps during data ingestion and baseline generation, before fraud analysis occurs. By correcting miscodings upfront, the system ensures data accuracy for subsequent detection processes without adding time pressure during critical real-time fraud analysis moments.
Data Source
AI summary
A method for detecting collusive transaction fraud includes: generating a merchant baseline including a transaction data baseline and a time series baseline; extracting time series data of the first merchant system; generating a first score and second score with a deep learning model; generating a first merchant risk score of the first merchant system based on the first and second scores; in response to determining that the first merchant risk score satisfies the threshold, determining a plurality of related entities related to the first merchant system; and classifying the first merchant system and at least one related entity of the plurality of related entities in a first group risk class based on at least one risk score of the at least one related entity.