CNN Breach Detection via Transaction Image Permutation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current breach detection systems face challenges due to data imbalance, requiring large volumes of samples for effective training, and often result in high rates of false positives and false negatives, leading to inefficiencies in fraud mitigation.
Innovation Solution
The use of convolutional neural networks (CNNs) to detect breach events by processing transaction data, specifically generating images from suspected fraudulent transactions and analyzing these images for patterns indicative of breaches.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional machine learning methods are used for breach detection, then large volumes of training samples are required, but the data is extremely imbalanced with only a few hundred confirmed breach events per year
Solution Approach 1:
The patent transforms tabular transaction data into image representations, enabling the use of CNNs that can automatically learn features from visual patterns. This dimensional transformation allows the system to detect breaches effectively even with limited training samples, as the image-based approach captures spatial and temporal relationships in transaction data that traditional tabular methods miss.
Solution Approach 2:
The patent replaces traditional machine learning algorithms (which require extensive manual feature engineering and large datasets) with convolutional neural networks that automatically extract features from image representations of transaction data. This substitution enables effective breach detection with limited training data by leveraging the CNN's ability to learn hierarchical patterns directly from the visualized transaction sequences.
2Reliability
If traditional breach detection systems are used, then false positives and false negatives increase, but this leads to computational inefficiencies in fraud mitigation processes
Solution Approach 1:
The patent implements a feedback mechanism where the CNN model continuously learns from confirmed breach events and adjusts its detection thresholds. The system incorporates feedback loops that refine the breach detection algorithm based on actual breach patterns observed in the data, reducing false positives and negatives over time while maintaining computational efficiency through optimized image processing and CNN inference.
3Loss of time
If current breach detection systems are used, then breach events are detected slowly and reactionarily, but a high number of transactions continue to occur at breached entities before detection
Solution Approach 1:
The patent transforms transaction data into image representations that capture temporal patterns and anomalies in real-time, enabling the CNN to detect breaches as they are occurring rather than after significant damage has been done. This preliminary detection capability allows security countermeasures to be activated immediately, preventing further fraudulent transactions while minimizing loss of time and reducing the overall transaction volume that occurs during active breaches.
Data Source
AI summary
Systems, methods, and computer program products are provided for breach detection using convolutional neural networks (CNNs). An example system includes a processor configured to generate a plurality of permuted images, each image comprising a field of points associated with suspected fraudulent transactions, an x-axis position of each point associated with a time, and a y-axis position of each point corresponding to a randomized index of a payment device. The processor is also configured to assign a breach likelihood score to each image using a CNN model. The processor is further configured to compare the breach likelihood score of each image to a threshold score. The processor is further configured to detect the breach event based on one or more breach likelihood scores satisfying the threshold score. The processor is further configured to, in response to detecting the breach event, decline transactions with the entity associated with the breach event.