Core Concepts
A novel collaborative learning model that enables blockchain nodes to effectively share their learned knowledge to improve the accuracy of detecting various cyberattacks, without the need to share their raw data.
Abstract
The key highlights and insights of this content are:
The authors developed a private blockchain network in their laboratory to generate a real dataset, named BNaT, for studying cyberattacks in blockchain networks. This is the first dataset obtained from a laboratory for this purpose.
The authors proposed a collaborative learning model where the fullnodes in the blockchain network act as learning nodes. Each learning node trains a deep belief network (DBN) model using its local data, and then shares the trained model with a centralized server. The centralized server aggregates the received models to create a global model, and sends it back to the learning nodes. This process is repeated until convergence.
The collaborative learning model allows the learning nodes to leverage knowledge from all nodes in the network to improve the accuracy of detecting various attacks, including Brute Password (BP), Denial of Service (DoS), Flooding of Transactions (FoT), and Man-in-the-Middle (MitM), without the need to share their raw data.
Extensive simulations and real-time experiments show that the proposed collaborative learning model can achieve an accuracy of up to 98.6% in detecting the considered cyberattacks, outperforming centralized learning and independent learning approaches.
The authors also provide insights on designing and implementing learning models in blockchain networks in practice, such as real-time monitoring and detecting attacks.
Stats
The accuracy of the proposed collaborative learning model can reach up to 98.6%.
The proposed model outperforms centralized learning and independent learning approaches in terms of accuracy, precision, and recall.
Quotes
"To the best of our knowledge, this is the first dataset that is synthesized in a laboratory for cyberattacks in a blockchain network."
"Our proposed model can achieve an accuracy of up to 98.6% in detecting cyberattacks in the considered network."