Transaction Anomaly Detection Using Merchant Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for assigning risk scores to merchant category codes for payment card transactions are inaccurate, time-consuming, and prone to errors due to the complexity and variability of merchant categories and sub-categories, leading to inefficient detection of unauthorized or ineligible transactions.
Innovation Solution
A system utilizing machine learning techniques, including an embedding model and an autoencoder, to generate similarity scores and out-of-pattern indices for transactions, enabling accurate and automated risk assessment at a granular level by analyzing transaction data structures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If manual scoring of merchant category codes is used to determine risk levels, then the process is simple to implement, but the accuracy and speed of anomaly detection deteriorates
Solution Approach 1:
The patent replaces manual scoring (mechanical human evaluation) with an automated machine learning system that uses embedding models and autoencoders to analyze transaction data, merchant categories, and spending patterns, thereby improving detection accuracy while maintaining ease of implementation through automation
Solution Approach 2:
The patent introduces embedding models as intermediaries that transform merchant category codes and transaction data into meaningful vector representations, enabling the system to capture semantic relationships and nuances that manual scoring cannot detect, thus improving accuracy without increasing operational complexity
2Ease of manufacture
If manual scoring of merchant category codes is used to determine risk levels, then the implementation is straightforward, but the processing speed deteriorates
Solution Approach 1:
The patent replaces time-consuming manual scoring with automated machine learning models that process transactions in real-time or near-real-time, dramatically improving processing speed while keeping the system easy to implement through standardized ML pipelines
Solution Approach 2:
The patent pre-trains embedding models on merchant category data and pre-processes transaction histories to create baseline spending patterns, enabling rapid anomaly detection during actual transaction processing without requiring complex real-time computations
3Device complexity
If manual scoring methods are used for risk assessment, then the system complexity is low, but the detection accuracy of unauthorized transactions deteriorates
Solution Approach 1:
The patent introduces embedding models as intermediaries that automatically capture semantic relationships between merchant categories and transaction contexts, enabling the system to achieve high detection accuracy through standardized ML components rather than complex custom rules
Solution Approach 2:
The patent uses autoencoders to create compressed representations (copies) of normal transaction patterns, allowing the system to detect anomalies by comparing new transactions against these learned patterns without requiring complex manual rule sets
4Measurement precision
If automated machine learning models are used for transaction analysis, then the detection accuracy improves, but the computational resources required increase
Solution Approach 1:
The patent extracts only the most relevant features from transaction data using embedding models that focus on semantic relationships between merchant categories and spending patterns, reducing the dimensionality and computational burden while maintaining high detection accuracy
Solution Approach 2:
The patent transforms discrete merchant category codes into continuous vector embeddings, enabling more efficient computational processing through mathematical operations on vectors rather than complex categorical comparisons, thereby reducing computational resource requirements
Data Source
AI summary
Disclosed herein are system, method, and computer program product embodiments for detecting an anomalous activity in a data structure. The method includes acquiring, by at least one processor, merchant category data and a plurality of authorized transactions, training an embedding model using the merchant category data. The embedding model receives an input merchant category for a transaction and generates a sentence embedding for the input merchant category. The method further comprises training an autoencoder using the plurality of authorized transactions. The autoencoder receives transaction data for the transaction and generates a similarity score for the transaction compared to the plurality of authorized transactions. The method further comprises generating a trained machine learning model that is configured to generate transaction scores and flag transactions based on the transaction scores. The trained machine learning model comprises the trained embedding model and the trained autoencoder.