MLChain: a privacy-preserving model learning framework using blockchain

In this work, we present a blockchain-based secure and flexible distributed privacy-preserving online model that helps in sharing key features of datasets across multiple organizations without violating the privacy of data. In our model, all members are encouraged to participate, discouraged to write fake data. Learning is carried out without sharing of raw data, and data sharing is immutable that improves prediction results of the data held by each member of an industry. We also propose a new consensus algorithm-Proof of Share for adding a valid transaction to the blockchain, thus preventing non participating members from reading any of the data shared by the peer and discouraging fake writes. We evaluated our model on 3, 5, and 10 members setup by applying decision tree, logistic regression, Gaussian naive Bayes, and support vector machine classifiers. The maximum increase of 26.9231% was observed in accuracy where results of a member's data were taken as baseline. F-beta(beta = 0.5) score increased by 0.4533 and F-1 score by 0.0800. The proposed model to the best of our knowledge is the only one that encourages all members to participate, rather than being passive listeners and discourages a member from forging results thus rendering it suitable for utilization in domains like health care, finance, education, etc. where data are unevenly split and secrecy of data and peers is required.