Master of Engineering Science
Electrical and Computer Engineering
Dr. Abdelkader Ouda
User authentication is considered a key factor in almost any software system and is often the first layer of security in the digital world. Authentication methods utilize one, or a combination of up to two, of the following factors: something you know, something you have and something you are. To prevent serious data breaches that have occurred using the traditional authentication methods, a fourth factor, something you do, that is being discussed among researchers; unfortunately, methods that rely on this fourth factor have problems of their own.
This thesis addresses the issues of the fourth authentication factor and proposes a data science solution for user authentication. The new solution is based on something you do and relies on analytic techniques to transfer Big data characteristics (volume, velocity and variety) into relevant security user profiles. Users’ information will be analyzed to create behavioral profiles. Just-in-time challenging questions are generated by these behavioral profiles, allowing an authentication on demand feature to be obtained. The proposed model assumes that the data is received from different sources. This data is analyzed using collaborative filtering (CF), a learning technique, that builds up knowledge by aggregating the collected users’ transaction data to identify information of security potential. Four use case scenarios were evaluated regarding the proposed model’s proof of concept. Additionally, a web based case study using MovieLens public dataset was implemented. Results show that the proposed model is successful as a proof of concept. The experiment confirms the potential of applying the proposed approach in real life as a new authentication method, leveraging the characteristics of Big data: volume, velocity and variety.
Ibrahim, Anas, "Data Science Solution for User Authentication" (2017). Electronic Thesis and Dissertation Repository. 4871.