Detecting Sybils in Peer-to-Peer Overlays Using Neural Networks and CAPTCHAs

Abstract

Over the years, peer-to-peer networks have emerged as one of the most popular file sharing medium over The Internet, capable of providing user anonymity to the clients if desired. However, modern P2P networks suffer from the bane of malicious entities we refer to as Sybils, which forge multiple identities to negatively influence or even control the entire network. This paper suggests a novel solution to eradicate the Sybil threat using a unique combination of neural networks and CAPTCHA. We capture common behavioral patterns of participating Sybil entities, in terms of certain quantitative variables, and ascertain their true identities by feeding these variables to a neural network, followed by sending CAPTCHA to the alleged entity ensuring a very high success rate in identifying malicious entities in the network. Network simulations have shown the proposed approach to be highly effective in countering the Sybil threat by giving a high degree of accuracy in detecting the malicious nodes.