Transaction Narrative Generation for Explainable AML Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Service providers face challenges in efficiently identifying and reviewing thousands of transactions for potential money laundering without a large review and compliance team, and agents struggle to understand the underlying data for flagged transactions.
Innovation Solution
A machine learning model trained with gradient boosting or random forest algorithms to detect prohibited transactions, accompanied by a narrative generator that explains the decision-making process, reducing false positives through iterative training and providing textual narratives for agent review.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual review of all transactions is performed, then detection accuracy is improved, but processing time and resource requirements increase significantly
Solution Approach 1:
The system segments the transaction review process into two stages: an automated machine learning model performs initial screening to flag suspicious transactions, and human agents only review the flagged subset. This segmentation maintains high detection accuracy while dramatically reducing the time and resources required, as agents don't need to manually review all transactions but only the ones identified as potentially suspicious by the model.
2Productivity
If automated filtering is used to reduce review workload, then processing speed is improved, but detection accuracy deteriorates due to increased false positives
Solution Approach 1:
The system implements feedback loops where human agents review flagged transactions and provide feedback on whether they are true positives or false positives. This feedback is used to iteratively retrain and improve the machine learning model, reducing false positives over time while maintaining high processing speed. The continuous feedback mechanism ensures the model learns from real-world outcomes and improves detection accuracy without sacrificing automation efficiency.
3Measurement precision
If complex analysis algorithms are applied, then detection accuracy is improved, but system complexity increases
Solution Approach 1:
The system introduces an explanatory layer that acts as an intermediary between the complex machine learning model and human agents. This layer generates natural language explanations that translate complex algorithmic decisions into understandable reasoning, allowing agents to comprehend why transactions were flagged without needing to understand the underlying complex algorithms. This intermediary maintains high detection accuracy while reducing the perceived system complexity for users.
4Measurement precision
If iterative model retraining is performed, then false positives are reduced, but computational resources and time increase
Solution Approach 1:
The system applies partial retraining by focusing computational resources on refining the model based on feedback from actual transaction reviews, rather than continuously retraining on all possible data. This targeted approach reduces false positives by concentrating computational effort on the most relevant learning opportunities from real-world outcomes, while avoiding excessive computational resource consumption that would occur with comprehensive continuous retraining.
Data Source
AI summary
There are provided systems and methods for a machine learning model and narrative generator for prohibited transaction detection and compliance. A service provider server, such as an electronic transaction processor, may generate a machine learning model using a supervised training technique, which may detect transactions that may be money laundering. The model may be iteratively trained by detecting flagged transactions and outputting those transactions to an agent for identification of false positives, which may be used to retrain the model. When outputting the flagged transactions, a narrative may be generated using an explainer graph and a machine learning prediction explainer that identifies the features of the transaction data that caused the transactions to be flagged. Further, once the model is trained additional transactions may be processed to determine whether the features of those transactions indicate prohibited behavior.