Reinforcement Learning-Based Anomaly Detection for Internet of Things Distributed Ledger Technology

Abstract

Distributed Ledger Technologies (DLT) are based on the Blockchain concept and have been specifically designed for enterprise-level devices with acceptable computing powers and network bandwidth. Direct Acyclic Graph (DAG) ledger(s) is a new form of DLT technology designed for Internet-of- Things (IoT) devices due to the nature of its disadvantages of the computing powers and limited network bandwidth. IOTA is a DAG-based Blockchain implementation for IoT applications that has gained an increased attention in recent years. One of the major concerns that is hindering for its wide adaptation is the security concerns. Many security attack occurrences against the IOTA such as parasite attacks, double spending, and DDoS to disrupt availability resources of the new ledger can become both widespread and disruptive. Existing security studies are ad-hoc and typically address a solution scheme for a specific security threat. In this paper, we present an adaptive Reinforcement-Learning (RL) approach to best classify the monitored resource consumption parameters of the DAG-based nodes or devices for any potential security anomaly detection. The aim is to create high accuracy security threat index that can be used to proactively defend the decentralized IOTA infrastructure and individual nodes against compromises. The performance evaluation results of this solution against DoS attacks are promising. The framework implementation derives a stochastic interpretation and output and the same time it converges deterministically.