A Heuristic Approach for Compromised Account Detection through Machine Learning

Abstract

Now-a-days the Online Social Network has become the day-to-day term. Almost everybody has the OSN account. With the increasing use of OSNs, the activity of spammers has also increased. Attackers compromise the legitimate users accounts to spread spam messages. Previously techniques like SybilRank, PCA, COMPA, spam filtering, honeypots etc. tried
to detect compromised accounts on OSN, but the major disadvantage was that they were unable to detect attack at run-time. Online social behavior of users can be studied and analyzed in order to form the social behavior profiles for individual user. It will be very difficult as well as costly for an attacker to fully imitate the behavior of legitimate user and thus lead to a mismatch in behavior, which will significantly tell the difference between legitimate access and compromised access. This paper uses FCM clustering for classifying the behavior of each user. HMM is used for determining the hidden states in the profile
comparison state. This paper eventually studies many more aspects of user social behavior in order to come up with more accurate and fast detection of compromised accounts of OSN.