Fraud Detection via Multivariate Payment Sequence Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current risk control systems for detecting financial fraud in payment transactions are passive and ineffective, failing to identify suspicious patterns across multiple payment instruments, leading to significant revenue loss and negative customer experience.
Innovation Solution
A computer-implemented method using a server system that generates multivariate payment transaction sequences from multiple payment instruments to predict and compare simulated and real transaction sequences via generative adversarial networks (GAN), determining fraudulent transactions based on deviations exceeding a threshold.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional passive fraud detection models are used, then the system structure is simple, but the fraud detection accuracy and timeliness deteriorate
Solution Approach 1:
The patent combines multiple payment instrument sequences into a unified multivariate sequence, integrating information from different cards held by the same user. This merging allows the system to detect fraud patterns that span across multiple instruments, significantly improving detection accuracy while maintaining a coherent system architecture through the GAN framework.
Solution Approach 2:
The patent transitions from analyzing univariate sequences (single payment instrument) to multivariate sequences (multiple payment instruments). This dimensional expansion enables the system to capture cross-instrument fraud patterns and user behavior correlations, enhancing detection precision without excessive complexity increase due to the structured GAN approach.
2Reliability
If fraud detection is performed in real-time using multiple payment instruments, then the fraud detection accuracy improves, but the computational complexity and processing time increase
Solution Approach 1:
The patent employs offline training of the GAN model using historical payment transaction data from multiple instruments. This preliminary action prepares the model in advance, enabling it to perform rapid real-time inference during actual fraud detection. The computationally intensive training phase is separated from the real-time detection phase, reducing online computational burden while maintaining high reliability.
Solution Approach 2:
The generator in the GAN creates simulated payment sequences that replicate real user spending patterns. These synthetic copies are used to train the discriminator, enabling the system to learn fraud detection without requiring excessive computational resources during actual operation. The copying mechanism allows efficient real-time detection by comparing against pre-learned patterns.
3Productivity
If individual payment transactions are analyzed in isolation, then the processing speed is fast, but the fraud detection capability deteriorates
Solution Approach 1:
The patent merges transaction data from multiple payment instruments into a unified multivariate sequence, enabling the detection of fraud patterns that involve coordinated use of different cards. This combining approach maintains processing efficiency by using the GAN framework to handle the integrated data stream, achieving both speed and precision.
Solution Approach 2:
The patent segments the analysis into distinct components: individual payment sequences from each instrument are extracted separately, then integrated into a multivariate sequence for comprehensive analysis. This segmentation allows efficient processing of individual transactions while capturing cross-instrument fraud patterns through the structured GAN architecture.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Embodiments provide methods and systems for detecting frauds in payment transactions made by payment instrument using spend patterns of multiple payment instruments associated with user (102). The method (700) performed by server system (108; 200) includes receiving (702) payment transaction data associated with first payment instrument including information of payment transaction performed at particular merchant (106a-c). Method (700) includes generating (704) multivariate payment transaction sequence associated with one or more second payment instruments of user (102). Method (700) includes predicting (706) simulated univariate payment transaction sequence associated with the first payment instrument based on first neural network model and the multivariate payment transaction sequence. Method (700) includes providing (708) simulated univariate payment transaction sequence and real univariate payment transaction sequence of first instrument to second neural network model. Method (700) includes determining (710) that payment transaction is fraudulent based, at least in part, on comparison of simulated univariate payment transaction sequence and real payment transaction sequence by the second neural network model.