User Flow Sequence Modeling for Early Website Fraud Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing fraud detection mechanisms in the financial sector, particularly for Account Takeover (ATO) fraud, fail to adequately consider the sequential flow of user interactions on websites, missing sophisticated fraud patterns that emerge from combined activities.
Innovation Solution
The use of sequential machine learning models, such as Hidden Markov Models (HMM), to analyze user flow data as a sequence of webpage transitions, converting it into vector representations, and classifying these sequences as fraudulent or legitimate, enabling early detection of ATO fraud.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional fraud detection mechanisms are used, then the detection process is simple and fast, but the detection precision is insufficient for sophisticated fraud patterns
Solution Approach 1:
The user flow data is segmented into discrete sequential steps representing individual webpage transitions or interactions. Each step is treated as a separate observation point in the sequence, allowing the system to analyze specific fraudulent patterns at each transition point while maintaining the overall contextual flow. This segmentation enables precise detection of anomalous transition patterns without requiring complex analysis of entire user sessions.
Solution Approach 2:
The patent transforms sequential user flow data into a new dimensional representation by creating state transition models and probability distributions. Instead of analyzing raw sequential data directly, the system maps transitions between user states into a probabilistic space where fraudulent patterns become distinguishable through statistical deviations. This dimensional transformation enables sophisticated fraud detection while maintaining computational efficiency.
2Measurement precision
If sequential machine learning models are used to analyze user flow data, then the detection precision improves, but the processing time and computational resources increase
Solution Approach 1:
The system performs preliminary actions by pre-computing state transition probabilities and establishing baseline models of normal user behavior during off-peak periods. These pre-computed models are stored and readily available for rapid comparison against incoming user flow data. When fraud detection is needed, the system only needs to compare current transitions against pre-established patterns, significantly reducing real-time processing requirements while maintaining high detection precision.
Solution Approach 2:
The patent applies partial action by focusing computational resources only on analyzing the sequential transition patterns that are most indicative of fraud, rather than processing all user interaction data equally. The system identifies and prioritizes critical transition points where fraudulent behavior is most likely to manifest, applying sophisticated sequential analysis only where needed while using simpler heuristics for other portions of the user flow, thus balancing detection precision with processing efficiency.
Data Source
AI summary
Mechanisms are provided for detecting fraudulent user flows associated with a website. User flow data, representing an interaction by a user with content of a website, is received and converted to a vector representation that represents a time series transition from one portion of website content to another of the website. The vector representation is input to a trained sequential machine learning computer model which generates a classification of the vector representation. A determination as to whether or not the user flow data represents a fraudulent user flow is made based on the classification. An output is generated that indicates whether or not the user flow is a fraudulent user flow based on the detection.