Wavelet-Based Transaction Fraud Detection System
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for transaction fraud detection in financial systems, such as rule-based determinations and traditional neural networks, face challenges with high computational resource requirements and latency, especially in handling large transaction volumes, leading to poor user experience and inefficiencies in real-time processing.
Innovation Solution
A data processing method that generates wavelets from transaction data, stores them in a key-value database, and outputs indicators for real-time processing, utilizing wavelet-based data processing to facilitate efficient and scalable transaction analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional neural networks are used for transaction fraud detection, then detection accuracy can be improved, but computational resource requirements and latency increase significantly
Solution Approach 1:
The patent transforms transaction data from traditional tabular format into wavelet-based time-frequency representations, changing the parameterization of the data. This transformation enables the use of simpler neural network architectures that operate on wavelet coefficients rather than raw transaction features, thereby reducing computational complexity while maintaining detection accuracy
Solution Approach 2:
The patent replaces complex deep learning architectures with a simplified neural network design that processes wavelet-transformed data. By substituting the data representation mechanism (using wavelets instead of traditional features), the system achieves comparable or superior detection performance with significantly reduced computational resources and latency
2Reliability
If complex neural networks are used to handle large transaction volumes, then detection capability is improved, but processing speed and user experience deteriorate
Solution Approach 1:
The patent applies wavelet transformation as a preprocessing step before neural network classification. This preliminary action of transforming transaction data into the wavelet domain extracts essential temporal and frequency features in advance, enabling the neural network to work with compressed, informative representations that require less computational processing during real-time transaction evaluation
Solution Approach 2:
By changing the data representation parameters from traditional transaction features to wavelet coefficients, the system reduces the dimensionality and complexity of the input data. This parameter transformation allows for faster processing speeds while maintaining the ability to detect fraud patterns in large transaction volumes
3Adaptability or versatility
If traditional financial indicators are used, then compatibility with existing systems is maintained, but performance with machine learning methods deteriorates
Solution Approach 1:
The patent creates a universal data representation framework using wavelets that can serve both traditional analytical methods and machine learning approaches. The wavelet transformation provides a multi-functional interface that maintains compatibility with existing financial systems while simultaneously optimizing performance for ML algorithms by providing time-frequency localized features that capture transaction patterns more effectively
Data Source
AI summary
A data processing method includes receiving input data associated with a plurality of transactions; generating, based on the input data, a plurality of wavelets corresponding to the plurality of transactions; storing the plurality of wavelets and corresponding keys associated with the wavelets in a key-value database; and outputting, based on the plurality of wavelets, one or more indicators.