Improving the performance of the Proof-of-Work Consensus Protocol Using Machine learning

Blockchain technology has proven to be secured and reliable technology by bringing security, trust and data integrity to distributed systems. It brought a new paradigm that helps in the existence of the cryptocurrency and eliminating the third party in a financial transaction. It has the potential of optimising, enhancing streamlining many processes outside the cryptocurrency and financial sector but the adoption of the technology is limited by the hindering performance issues. These issues are mostly around the Proof-of-Work (PoW) consensus protocol that is used by Bitcoin and Ethereum and referred to as the most secured and decentralised protocol thus, the most reliable. Unfortunately, the protocol suffers a performance degrade with the increasing size and number of transactions because of its complexity. Many industries, researchers and organisation have been working on addressing these issues but most of the attempts result in facing another issue referred to as the scalability issue; having to trade off one of security or decentralisation to get speed. Other solution such as Bitcoin lighting network improved the transaction throughput of the Bitcoin without technically addressing the issue on the blockchain, therefore, didn't face the scalability issue. This paper presents a research work of a novel approach that propose using machine learning techniques to improve the performance by enhancing the mining efficiency of the protocol. It uses the Ethereum network as a case study. The paper also aims to compare the predictive accuracy and speed of some machine learning regression models against the traditional mining method starting from the Linear regression. The objective is to determine whether using machine learning in the mining process offers a faster way of achieving consensus, therefore improving performance by reducing the time and energy consumption of the protocol without sacrificing security or decentralisation. The proposed model results in improved accuracy and faster consensus from the early experiments.