Dynamic Variable Profiling for Synthetic Identity Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing use of synthetic identities poses challenges in quickly and efficiently identifying unauthorized data access and activity, as authentication data can be obtained by unauthorized users, making it difficult for enterprise organizations to detect anomalies.
Innovation Solution
A computing platform uses machine learning to generate dynamic variable profiles based on historical user data from internal and external sources, analyzing user-specific data to detect potential unauthorized activity by comparing it against these profiles, and implementing mitigating actions when anomalies are detected.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional authentication methods are used, then user access can be verified, but unauthorized users can still compromise authentication data and create synthetic identities
Solution Approach 1:
The patent introduces an intermediary machine learning model that acts as a mediator between authentication data and access decisions. This model analyzes multiple variables and patterns to detect synthetic identities, preventing unauthorized access even when authentication data is compromised. The intermediary layer adds an additional security mechanism that goes beyond traditional authentication.
Solution Approach 2:
The system performs preliminary analysis by training the machine learning model on historical data before actual authentication events occur. The model pre-establishes patterns of legitimate user behavior and identifies characteristics of synthetic identities in advance, enabling proactive detection rather than reactive response to compromised credentials.
2Measurement precision
If machine learning models are trained on historical user data, then detection accuracy improves, but system complexity and computational resources increase
Solution Approach 1:
The patent segments the detection system into distinct components: historical data collection, model training phase, and real-time inference phase. This segmentation allows the complex training process to be separated from operational requirements, enabling the system to achieve high detection accuracy while managing complexity through modular architecture.
Solution Approach 2:
The complex task of pattern recognition is performed in advance during the model training phase using historical data. Once trained, the model captures complex patterns and simplifies real-time detection to straightforward pattern matching, reducing computational complexity during operational use while maintaining high detection accuracy.
3Reliability
If real-time evaluation of user data is performed, then unauthorized activity can be detected quickly, but processing time and computational load increase
Solution Approach 1:
The machine learning model pre-processes and learns from historical user data during an offline training phase, capturing complex patterns and relationships in advance. This preliminary action transfers computational burden from real-time operation to offline training, enabling fast real-time evaluation with minimal processing time while maintaining high detection reliability.
Solution Approach 2:
The trained model creates a simplified representation or copy of complex user behavior patterns that can be quickly evaluated in real-time. Instead of performing complex analysis on raw data during authentication, the system uses the trained model's learned patterns, which are essentially compressed copies of behavioral characteristics, enabling rapid detection.
Data Source
AI summary
Arrangements for dynamic variable determination and labeling are provided. In some aspects, a computing platform may receive historical user data from a plurality of data sources. The computing platform may train, using the historical user data, a machine learning model to generate a plurality of dynamic variable profiles and evaluate data to detect potential unauthorized activity. One or more dynamic variable profiles of the generated plurality of dynamic variable profiles may be associated with a user. User specific data may be received and may include user identifying data and a request for a user event. The user specific data may be input to the machine learning model and, upon execution of the model, the model may output a determination of whether an anomaly exists in the user specific data. If an anomaly is detected, a mitigating action may be identified and transmitted to one or more computing devices for execution.