Two-Phase Fraud Detection Using Homomorphic Encryption
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In high-throughput transaction processing environments, such as credit card authorization, predicting fraud is challenging due to imbalanced data sets and the need for real-time detection, which is complicated by the rarity of fraudulent activity and the cost of false positives.
Innovation Solution
A two-phase inferencing method is implemented, where a first machine learning model is trained on non-private transaction data and performs initial evaluations in the clear, while a second model, trained on both transaction and private user data, performs a more secure evaluation on encrypted data using homomorphic encryption, only flagging transactions as potentially fraudulent if both models agree.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If homomorphic encryption is used to perform privacy-preserving inferencing on high-accuracy models trained with private user data, then data privacy is protected, but the computational complexity and processing time become unacceptable for high-volume transaction environments
Solution Approach 1:
The inferencing process is segmented into two distinct phases: a first phase using a model trained only on non-private data for initial filtering, and a second phase using a model trained on private data with homomorphic encryption for final verification. This segmentation allows the computationally intensive encrypted operations to be applied only to a small subset of transactions that pass the first filter, thereby maintaining privacy protection while achieving acceptable throughput for high-volume environments
2Measurement precision
If a single high-accuracy model trained on private user data is used for fraud detection, then detection accuracy is improved, but the computational intensity on encrypted data becomes prohibitive for real-time processing
Solution Approach 1:
Instead of applying the computationally intensive homomorphic encryption-based model to all transactions, the system applies it partially only to transactions that are flagged by the first model. This partial action approach maintains high detection accuracy for suspicious transactions while dramatically reducing the overall computational intensity and energy consumption across the entire transaction stream
3Reliability
If homomorphic encryption is applied to all transactions in high-volume data streams, then privacy is preserved, but the processing time and computational resources required become unmanageable
Solution Approach 1:
The system performs preliminary filtering using a first model trained on non-private data before applying homomorphic encryption. This preliminary action identifies and filters out the majority of legitimate transactions, ensuring that privacy-preserving encrypted processing is applied only to the small fraction of transactions that require further scrutiny, thereby preserving privacy while minimizing processing time loss
Data Source
AI summary
A method, apparatus and computer program product that provides multi-phase privacy-preserving inferencing in a high throughput data environment, e.g., to facilitate fraud prediction, detection and prevention. In one embodiment, two (2) machine learning models are used, a first model that is trained in the clear on first transaction data, and a second model that is trained in the clear but on the first transaction data, and user data. The first model is used to perform inferencing in the clear on the high throughput received data. In this manner, the first model provides a first level evaluation of whether a particular transaction might be fraudulent. If a transaction is flagged in this first phase, a second more secure inference is then carried out using the second model. The inferencing performed by the second model is done on homomorphically encrypted data. Thus, only those transactions marked by the first model are passed to the second model for secure evaluation.