The immense popularity of social networks has rendered them vulnerable to security threats such as Sybil attacks. In Sybil attacks, attackers send several friend requests from fake identities, and some of the genuine users might accept them unknowingly and become victims of Sybil attacks. Once genuine users accept friend requests from Sybil identities, attackers can spam, phish, and conduct other harmful actions in the target networks. Though genuine users have links to Sybil users unknowingly, they do less interaction and have low strength of relationship with them. Thus, the relationship strength measure can be useful to detect a Sybil attack. Considering the relationship strength measure, we propose a behavior-based and graph-based hybrid approach to detect a Sybil attack in Online Social Networks (OSNs). We identify the behavior features, use them to measure the relationship strength, and identify attack edges. We use the graph-based feature(betweenness-centrality) to leverage attack-edge identification and detect Sybil nodes. We have evaluated our scheme using real-world datasets, and experimental results validate the proposed scheme's effectiveness. [Display omitted] • A hybrid scheme is suggested to detect Sybil attacks in online social networks. • We identify and apply behavior-based important features to identify attack edges. • An important graph-based structural feature is employed to detect Sybil nodes. • The proposed scheme is effective and outperforms current approaches. [ABSTRACT FROM AUTHOR]