Optimization of consumer data protection algorithms and privacy protection technologies in the digital age

In the digital age, consumer data is widely used. In this article, a multi-level privacy protection algorithm combining Differential Privacy (DP) mechanism and Federated Learning (FL) framework is presented, aiming at improving the privacy security and practicability of model training. By adopting local feature dimensionality reduction, two-stage differential noise injection and edge aggregation strategy, a new architecture considering both privacy budget and model performance is constructed. The experiment is based on Adult, Criteo and e-commerce data sets, and the performance of the model is evaluated under different privacy budget conditions. The results show that when ε=1.0, the accuracy of the proposed method can reach 87.2%, which is 6.3% higher than the traditional FedDP algorithm. At the same time, in the member reasoning attack scenario, the attack recognition rate is reduced to 48%, which greatly improves the system security. The research shows that the algorithm effectively meets the requirements of data anonymity and minimization in regulations such as GDPR on the basis of ensuring data availability.