A federated learning-based selection and incentive system using blockchain technology

Machine learning algorithms are powerful tools for analyzing data with several observations approximately equal to the number of predictors. However, the privacy of data owners may be revealed during the processes of analysis and mining in a distributed scenario. Today, federated learning is employed as the best paradigm for collaborative model training without disclosing the privacy of the data owners. Unfortunately, efficient client selection and incentive mechanisms need to provide for encouraging data sharing and analysis for constrained and non-constrained devices. Furthermore, trust in the system must be considered. To this end, this study proposes a federated blockchain-based incentive and selection mechanism for a federated learning system. Clients are selected using support vector machines (SVM), while the accuracy of SVM is improved by recursive feature elimination (RFE). A real-time incentive is provided to clients for collaborative learning using deep Q reinforcement learning, and an optimal incentive allocation policy is derived using the Markov decision process (MDP) framework. For miners’ selection, a proof of utility consensus is proposed using a sixteen-round addition game. Extensive simulations are conducted to evaluate the efficiency of the proposed system model. The performance of the proposed system is determined by its optimal statistical utility, system utility, and client utility, respectively. From the experimental results, the proposed SVM-RFE model outperform the existing algorithms. Additionally, security analysis is performed, which shows that the proposed system is safe against background knowledge attacks.