Machine Learning Fraud Scoring With Streaming and Batch Enrichment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current fraud detection systems struggle to identify fraudulent financial transactions, particularly in mobile and open banking environments, due to inadequate infrastructure and the rise of first-party fraud.
Innovation Solution
A deposit fraud machine learning model is employed to generate alerts for suspicious transactions by assigning risk indicators based on transaction factors, allowing for immediate detection and flagging of potentially fraudulent activities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional fraud detection systems are used, then infrastructure complexity is reduced, but detection precision and reliability deteriorate due to inability to identify sophisticated fraud patterns
Solution Approach 1:
The patent introduces machine learning models as intermediary components between transaction data and fraud detection decisions. These models process and analyze transaction patterns, user behavior, and device information to generate fraud risk scores, serving as a mediator that transforms raw data into actionable intelligence without requiring complex rule-based systems
Solution Approach 2:
The system dynamically changes detection parameters by adjusting fraud risk thresholds, model confidence levels, and alert priorities based on transaction context, historical data, and emerging fraud patterns. This allows the system to adapt its detection sensitivity without requiring complete system redesign
2Speed
If real-time fraud detection is implemented, then response speed improves, but processing time and computational resources increase
Solution Approach 1:
The system performs preliminary actions by pre-processing transaction data, pre-training machine learning models on historical fraud patterns, and pre-establishing risk thresholds before actual fraud detection occurs. This preparation enables rapid real-time analysis without extensive processing during critical detection moments
Solution Approach 2:
The system implements selective skipping by bypassing full model analysis for low-risk transactions and focusing computational resources only on suspicious activities. This allows the system to process high-volume transactions quickly while maintaining thorough analysis where needed
3Reliability
If comprehensive transaction monitoring is applied, then detection coverage improves, but false positive rate increases leading to more alerts requiring review
Solution Approach 1:
The patent applies local quality by implementing context-specific detection rules and thresholds for different transaction types, users, and risk categories. Rather than uniform monitoring, the system tailors detection sensitivity to local characteristics such as transaction amount, user history, device reputation, and geographic location, reducing false positives while maintaining comprehensive coverage
Solution Approach 2:
The system incorporates feedback mechanisms where detection results, analyst reviews, and confirmed fraud cases are fed back into the machine learning models to continuously refine detection accuracy. This feedback loop reduces false positives over time by learning from actual outcomes and adjusting detection parameters accordingly
4Measurement precision
If machine learning models are deployed, then detection accuracy improves, but system complexity and training requirements increase
Solution Approach 1:
The system implements self-service by enabling automatic model training, hyperparameter optimization, and performance monitoring without requiring extensive manual intervention. The machine learning models continuously learn from incoming data and automatically adjust to new fraud patterns, reducing the complexity of model maintenance and deployment
Data Source
AI summary
Systems and methods are disclosed for identifying unauthorized activity in a computing environment using a machine learning model. In disclosed embodiments, a computing system determines an initial unauthorized activity score for a processed action; and receives streaming data and batch data corresponding to a processed action occurring on a transaction channel computing device. The system processes the streaming and batch data to generate respective engineered features, which are combined into an integrated feature. An initial unauthorized activity score is determined for the processed action, and an updated score is calculated based on the integrated feature. When the updated score differs from the initial score by a predetermined enrichment threshold, a notification is transmitted to a user interface device. The system supports real-time enrichment of the processed action by appending additional input data, including engineered streaming and batch features, to improve detection accuracy.