Healthcare Data Protection using Federated Learning Technology
B. Gobinath, T.K. Shanmugam · 2023
Federated learning makes it possible to train models collectively without distributing raw data. Recent assaults, however, show that merely preserving data location during training procedures does not give substantial user privacy. Instead, there is a need for a federated learning system that can avoid generalization over both the training-related messages and the final trained model, while also preserving the model’s appropriate forecasting ability. Current federated learning systems either use differential privacy, which can result in low accuracy provided there are a large number of participants, or functional encryption which is susceptible to generalization. In this research, an alternate method, Polymorphic Cipher Functional Encryption for balancing the limitations of these methods is proposed. The increase of noise infusion is minimized as the number of people involved grows without compromising privacy and while maintaining a reasonable rate of trust by integrating differential privacy with functional encryption. Further, the proposed system is evaluated using Deep Convolutional Neural Networks model by applying it to diabetes dataset. The suggested technique proves to be a modular strategy that guards against generalization hazards and generates models with a high degree of accuracy.