Detection of Sybil Attacks in Online Social Networks

Abstract

Online social networks (OSNs) like Facebook and Twitter have become increasingly popular, offering a wide range of virtual interaction techniques and real-world social connections. The users of social media increasing day-by-day, these networks are expected to expand as mobile device usage and mobile social networks become more popular. However, Sybil attacks are a growing security issue in OSNs, where attackers use various methods to target large populations, creating fake identities and accessing networks. We carried out the comparison between ”Content based”, and ”user behavior based and graph based hybrid approach” to detect the Sybil’s attacks in OSNs. For user behavior and graph based approach, we extract the features from dataset and then find the behavior-similarities between the nodes and find the betweenness centrality between nodes. Behavior similarities values assigned to the edges as weights. and betweenness centrality value assigned to nodes. To detect Sybil nodes we define the threshold 20%, the behavior similarities value less then the threshold value, the edge become a Sybil edge whereas the to detect sybil nodes, the value of betweenness centrality less the threshold, the node become a Sybil node. For content based approach, we detect Sybil nodes in OSNs.Content based approach utilizes Machine Learning algorithms suc as Naive Basiyan, Multi Laye precepton, KNN, Random Forest, Logistic regression, SVC and ADA Boost. The Random Forest Algorithms perform well from all these algorithms to identifies Sybil nodes with high accuracy, precision and recall. we compared these two approaches to identify the attack edges and sybil nodes in social networks and the results revealed that the user behavior based and graph based hybrid technique perform well in term to identify the Sybil attacks with accuracy of 98.87%.