Research on Financial Transaction Data Protection and Intelligent Risk Assessment Based on Differential Privacy

This study explores the application of differential privacy techniques in the protection of financial transaction data. Utilizing the “Credit Card Fraud Detection” dataset, we implemented the Laplace and Exponential mechanisms to ensure the privacy of sensitive data features such as transaction amount and transaction time. We then employed a Long Short-Term Memory (LSTM) model trained with Differential Privacy Stochastic Gradient Descent (DP-SGD) to evaluate the impact of privacy-preserving measures on model performance. The results show that despite the introduction of differential privacy mechanisms, the model's performance metrics, including accuracy, recall, and F1-Score, remained at high levels, validating our approach's effectiveness in preserving data privacy while maintaining utility and efficiency. The innovations of this study lie in the comprehensive application of the Laplace and Exponential mechanisms, the incorporation of the DP-SGD algorithm in LSTM model training, ensuring privacy protection throughout the entire process from data preprocessing to model training. Additionally, we utilized multiple evaluation metrics and visualization tools to demonstrate the impact of differential privacy techniques on financial data in detail, providing new insights and methods for the field of financial data privacy protection. Future research may further optimize the parameters of differential privacy mechanisms and apply the method to various financial datasets with different characteristics to validate its generalizability and robustness.