Biometrics continuous authentication for detecting compromised accounts on Facebook

Other Title(s)

المصادقة المستمرة باستخدام القياسات الحيوية لكشف الحسابات المخترقة على ال Facebook

Dissertant

al-Qatawinah, Rawan Nashat

Thesis advisor

al-Madi, Nailah
al-Qatawinah, Jafar

University

Princess Sumaya University for Technology

Faculty

King Hussein Faculty for Computing Sciences

Department

Information Systems Security and Digital Criminology

University Country

Jordan

Degree

Master

Degree Date

2018

English Abstract

Online Social Networks have become one of the key channels for social communication.

One of the main features that characterize most online social networks is that they are profile based services where a user needs to create an account to start using their services.

Such accounts or profiles are subject to several security and privacy threats.

Account breakthrough represents one of the most critical security threats to users as it violates account owner privacy and an imposter can exploit account information and pre-established connections to use it in a bad and abusive manner.

This thesis addresses the use of behavioral biometrics to detect intrusions on a Facebook account and authenticate users on a continuous manner.

In particular, keystroke dynamics features in addition to social interaction patterns are used to characterize user behavior over Facebook.

The efficiency of keystroke dynamics features are validated by collecting and analyzing real user keystrokes during their normal use of Facebook.

While social behavioral features are validated through collecting and analyzing same users Facebook activity logs.

Four well known machine learning classifiers including (Decision trees, Random Forest, Multi-Layer Perceptron neural networks, and Naïve Bayes) are applied and tested on collected data to evaluate the capability of built models in discriminating legitimate users from impostors.

Experimental performance results for both keystroke dynamics features as well as social behavioral features indicate that built classification models are able to discriminate a legitimate user from an impostor with 90.8% accuracy rate obtained using keystrokes based models and 99% accuracy rate obtained using social behavior based models.

It has been shown through conducting experiments that classification models perform better for detecting impostor than for continuous authentication approach, with precision values that range between 90% and 93% respectively, and recall values that range between 92% and 97% for the keystrokes based models.

The same inference with precision and recall results for the social behavior based models that ranges between 94% and 100% precision, and 93% and 100% recall.

Main Subjects

Information Technology and Computer Science

Topics

• Intrusion detection systems(Computer security)
• Facebook(Electronic resource)

No. of Pages

70

Table of Contents

• APA
• MLA
• AMA

Language

English

Data Type

Arab Theses

Record ID

BIM-833375