A Machine Learning Consensus Based Light-Weight Blockchain Architecture for Internet of Things

Abstract

Blockchain is considered as an important technique for maintaining the integrity of data in enterprises. However, the computationally intensive proof-of-work forms a major bottleneck in adopting blockchains to energy-constrained environments such as Internet of Things (IoT). In this paper, we propose a Machine learning Consensus based Light-weight Blockchain (MCLB) for resource-constrained edge devices in IoT to detect malicious data besides providing the consensus for maintaining the integrity of data. Our light-weight approach reduces the overhead by eliminating the nonce of traditional blockchains, thereby bringing down the delay in arriving at a consensus. Each edge device is equipped with a machine learning algorithm to classify data and the predicted classifications are used to arrive at the consensus. The use of different machine learning algorithms at the edge nodes improves the robustness of our system. Our framework is realized on a sensor network consisting of four edge devices and the results show that MCLB outperforms existing blockchain models in terms of computational delay, communication overhead, and block addition delay.