Ethereum Eclipse Attack Detection Based on Multi-Head Attention Mechanism with Bi-LSTM

Abstract

The rapid development of blockchain technology and the rise of Ethereum as its representative platform has triggered a wide range of research and applications. However, this development is also accompanied by new security challenges, among which the Eclipse attack is one of the significant security threats currently facing Ethereum networks. In response to these challenges, we propose an Ethereum Eclipse attack detection method based on a multi-head attention mechanism with Bi-LSTM. This approach utilizes the Bi-LSTM model and multi-head attention mechanism to process time-series data, capturing and focusing on the features most relevant to the Eclipse attack for accurate identification. Additionally, we employ PCA and UMAP dimensionality reduction techniques in data preprocessing to enhance processing efficiency. Experimental results demonstrate that this method distinguishes regular traffic from attack traffic more accurately. Compared to the existing random forest method, our detection approach based on a multi-head attention mechanism with Bi-LSTM achieves a higher detection rate and lower false alarm rate, highlighting its effectiveness in addressing Ethereum network security.