Security Enhanced Federated Learning Approach using Blockchain

Abstract

In traditional machine learning approach, data gathered from all the edge devices are sent to centralized server for training and prediction of the output. In the centralized approach, user has to compromise on the data privacy and integrity in sharing their own data to centralized server. To overcome this issue federated machine learning approach was introduced, in which model and data are decentralized and the machine learning model will be trained on the data in local devices and parameters will be sent to cloud server for consensus change, enhancing the data privacy of the users. But still authentication of the nodes to cloud server and vice versa is a major concern to be addressed in federated machine learning as malicious nodes can impersonate as authenticated node and communicate to cloud server. In the proposed model, node authentication is implemented using Ethereum based blockchain with smart contracts thereby enhancing security of Federated machine learning approach. The efficiency of the node authentication is measured and compared with machine learning algorithms which achieves 99% accuracy.