الفهرس 
INTRODUCTIONOnly 14 pages are availabe for public view
 |  | from | 156 |  |  |
| | | from | 156 | | |
Abstract
Nowadays, touchscreen mobile devices make up a larger share in the market. Users often use these devices to store personal and sensitive data. This necessitate to find more effective and robust methods to continuously authenticate touch-based mobile device users. In this thesis, we propose two levels of behavioral touch features that can be extracted from raw touchscreen logs and demonstrate that different users populate distinct subspaces of this feature space. in an experiment designed to test how this behavioral pattern exhibits consistency over time, we collected touch data from users interacting with a touch-based tablet device using the most common gestures, i.e., scrolling in all directions; up, down, right and left, tapping, zoom-in and out etc..,.
A classification framework is proposed to learn the touch behavior of a user during an enrollment phase and is able afterwards to authenticate users by monitoring their behavior in performing input touch actions. Two models of features are built from two different perspectives to study which model will give us the highest accuracy in classification; the low level features model obtained through the stoke-level data or the high level abstracted features model obtained through the session-level data. In building these models, two different methods for features selection and data classification are; weighted features and principle component analysis (PCA) to determine the most efficient feature selection method. Moreover, two different classification algorithms were used; artificial neural networks (ANN) and support vector machines (SVM). Thus, we have four possibilities from those combinations in each model.