Improving Security with Federated Learning

Data leakage is the intentional or unintended transmission of stable or personal data to outside recipient. Such leakage in mobile community increases the chance of compilation. Hence encryption and storage of the secure data must be accomplished by usage of a few techniques. Federated learning (FL), which falls under distributed machine learning, helps preserve clients’ private data on various device as the centralized model receives only weight updates. Sensitive private data is open for access by analyzing submitted attributes from clients using techniques like weights developed in deep neural networks. To effectively preserve statistics from leakage, this study analyzes a novel framework using differential privacy (DP), in which synthetic noises are provided to parameters on the customers' side prior to aggregation, FLAGnoise (FL with noise aggregated).The system analyses the dataset consisting of information about the client. Federated learning with Advanced Encryption Standard (AES) algorithm and Differential privacy is then applied. It is found that the Federated learning model have better privacy than the Differential privacy model and gives the accuracy of 97.3%.