bootstrap modal popup

        

CYBER SECURITY 

MACHINE LEARNING FOR DETECTING ANOMOLOUS ACCOUNT ACTIVITY

Mobirise

Social media accounts are a valuable source of information to hackers and malicious users. Statistics state that more than 6,00,000 social network accounts are compromised every day. The said count throws a huge concern over the security and privacy aspect of these platforms and thus calls for the development of efficient detection techniques that can warn users of probable compromise of their accounts at the earliest.

Mobirise themes are based on Bootstrap 3 and Bootstrap 4 - most powerful mobile first framework. Now, even if you're not code-savvy, you can be a part of an exciting growing bootstrap community.

Choose from the large selection of latest pre-made blocks - full-screen intro, bootstrap carousel, content slider, responsive image gallery with lightbox, parallax scrolling, video backgrounds, hamburger menu, sticky header and more.

Sites made with Mobirise are 100% mobile-friendly according the latest Google Test and Google loves those websites (officially)!

Mobirise themes are based on Bootstrap 3 and Bootstrap 4 - most powerful mobile first framework. Now, even if you're not code-savvy, you can be a part of an exciting growing bootstrap community.

Choose from the large selection of latest pre-made blocks - full-screen intro, bootstrap carousel, content slider, responsive image gallery with lightbox, parallax scrolling, video backgrounds, hamburger menu, sticky header and more.

Working in this direction an anomaly detection system is being developed that builds customized profiles of normal activity for each user. The regular consistent patterns of a user are profiled and any upcoming information is compared against this profile. Genuine users inadvertently adhere to the regular behavior pattern. But with an imposter exploiting the profile for his own interest, this behavior pattern often gets an automatic strict violation. This deviation in behavior is marked as a point of compromise. This in-turn makes it difficult for an attacker to decide on activities to perform in order to evade detection.

  It also helps to detect the deviation as soon as an anomalous activity is performed.  To automate the process, machine learning techniques are deployed for automatic binary classification.


The Figure above illustrates the machine learning based workflow to detect compromised social network(facebook, twitter etc.) accounts. For an unknown sample, the task is to determine whether the activity under consideration has been performed by the concerned user Ui or not. For that to happen, the process can be divided into three steps: 


  1. Firstly, the relevant features are extracted and profiled from the existing data samples of a user. Each user must have enough of data samples so as to have sufficient amount of labeled training data.
  2. Same set of features extracted from the unknown sample are compared against the learned patterns. Comparing with the already learned patterns of both positive (User Ui) and the negative class (NotUser Ui), the class with which the feature values of the unknown sample match the most are predicted as the label. If the feature value does not match the positive class, it is counted as an anomalous behavior and the account is detected to be probably compromised. It is noteworthy that according to Pariser’s Filter theory, every user unknowingly build his own bubble space based on his interest and search patterns. Hence, being ingenuous he usually keeps living in his own social space with the same like minded people. As a result his postings, topic of interests and social network usage build up a unique pattern which he inadvertently follows. Accordingly, each independent user is expected to maintain consistency in different set of features. Hence, unlike other domains, here the classifiers are trained and tested independently for each respective user because of the consistency maintained by each user on his/her own behavioral profile.
  3. For incorporating the dynamics and changing behavior of users, an incremental learning process is adopted by continuously updating the training data with new patterns.

Address 

University Institute of Engineering & Tech, Panjab University, South Campus, Sector-25, Chandigarh 160036 India

Contact 

Email:  maivriklabs@gmail.com 

CONTACT PAGE

Links 

Panjab University      
UIET    
Centre of SKill Devlopment & Entreprenureship  (CSDE)

Feedback 

Please send us your ideas, bug reports, suggestions !             Feedback would be appreciated.