Detection Of Insider Attacks In Block Chain Network Using The Trusted Two Way Intrusion Detection System

TL;DR

This paper proposes a trusted two-way intrusion detection system for blockchain networks using a hierarchical fuzzy trust evaluation and a deep learning SOSN model, improving detection accuracy of insider attacks.

Contribution

It introduces a novel combination of fuzzy trust evaluation and deep learning for insider attack detection in blockchain, with implementation and validation in MATLAB.

Findings

Enhanced detection accuracy with higher Precision and Recall.

Better performance over existing methods in F-Score and overhead.

Effective identification of malicious transactions in blockchain environments.

Abstract

For data privacy, system reliability, and security, Blockchain technologies have become more popular in recent years. Despite its usefulness, the blockchain is vulnerable to cyber assaults; for example, in January 2019 a 51% attack on Ethereum Classic successfully exposed flaws in the platform's security. From a statistical point of view, attacks represent a highly unusual occurrence that deviates significantly from the norm. Blockchain attack detection may benefit from Deep Learning, a field of study whose aim is to discover insights, patterns, and anomalies within massive data repositories. In this work, we define an trusted two way intrusion detection system based on a Hierarchical weighed fuzzy algorithm and self-organized stacked network (SOSN) deep learning model, that is trained exploiting aggregate information extracted by monitoring blockchain activities. Here initially the smart contract handles the node authentication. The purpose of authenticating the node is to ensure that only specific nodes can submit and retrieve the information. We implement Hierarchical weighed fuzzy algorithm to evaluate the trust ability of the transaction nodes. Then the transaction verification step ensures that all malicious transactions or activities on the submitted transaction by self-organized stacked network deep learning model. The whole experimentation was carried out under matlab environment. Extensive experimental results confirm that our suggested detection method has better performance over important indicators such as Precision, Recall, F-Score, overhead.