ML Threat Score Workflow for Fraud Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current digital fraud and abuse detection technologies lack accuracy and real-time response capabilities, failing to effectively detect new threats and automatically evolve to neutralize them.
Innovation Solution
A machine learning-informed automated verification system that uses feature extraction and threat scoring models to predict threat scores, which are evaluated against decisioning workflows to inform action on online activities, including automated verification and disposal decisions, with a multi-stage synchronous process flow and verification thresholds to determine appropriate responses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional fraud detection technologies are used, then detection coverage is provided, but detection accuracy and real-time response capability are insufficient
Solution Approach 1:
The system segments the fraud detection process into distinct stages: feature extraction, threat scoring, verification decisioning, and automated verification. Each stage is handled by specialized components (feature extraction systems, threat scoring machine learning models, verification workflow) that can operate independently and in parallel, enabling both high accuracy and real-time response
Solution Approach 2:
The system dynamically adjusts the verification process based on the computed threat score. When threat scores exceed verification thresholds, automated verification is triggered; otherwise, activities proceed normally. This dynamic adaptation enables real-time response while maintaining detection accuracy through context-aware decision-making
2Adaptability or versatility
If existing detection technologies are implemented, then some fraud detection is provided, but the capability to detect new threats and automatically evolve is lacking
Solution Approach 1:
The system incorporates feedback loops where verification results and threat patterns are fed back into the machine learning models for continuous training and improvement. This enables the system to automatically evolve and detect new threat types without manual intervention, enhancing adaptability while managing complexity through automated learning processes
Solution Approach 2:
The machine learning models automatically update and improve their threat detection capabilities through self-training on new data patterns. The system performs self-service evolution by continuously learning from verified fraud cases and emerging threat patterns, reducing the need for manual system updates while improving detection of new threats
3Productivity
If manual verification processes are used, then verification accuracy is maintained, but response time and automation level are reduced
Solution Approach 1:
The system introduces an automated verification intermediary that acts as a mediator between threat detection and manual review. This intermediary component automatically performs verification tasks using machine learning models and only escalates uncertain cases to human reviewers, thereby improving response time while maintaining appropriate automation levels for different verification scenarios
Data Source
AI summary
A system and method for a machine learning-based score driven automated verification of a target event includes: receiving a threat verification request; extracting a corpus of threat features; predicting the machine learning-based threat score; evaluating the machine learning-based threat score against distinct stages of an automated disposal decisioning workflow; computing the activity disposal decision, wherein the activity disposal decision informs an action to allow or to disallow the target online activity; receiving the machine learning-based threat score as input into an automated verification workflow; computing whether an automated verification of the target online activity is required or not based on an evaluation of the machine learning-based threat score against distinct verification decisioning criteria of the automated verification workflow; automatically executing the automated verification of the target online activity and exposing results of the automated verification to the subscriber for allowing or for disallowing the target online activity.