Continuous Authentication of Smartphones using a Convolutional Neural Network Model

Current smartphone security is not up to par with current smartphones. The main technology of today is single authentication, meaning that once the smartphone is unlocked it is prone to attacks such as smudging where hackers can steal the password information of the smartphone. To deal with this issue, researchers have come up with the idea of continuous authentication, which can authenticate the user while they are using the smartphone. The most common way currently is to use biometrics of the user such as their walking pattern, hand movement, orientation, and grasp. Sensors built into the smartphone including the accelerometer, gyroscope, and magnetometer capture data readings for the smartphone which are then sent to an authentication algorithm to confirm the identity of the user. The goal is to be able to run an authentication algorithm in the background of the smartphone and that cannot be detected by the user. In this project, the researchers focus on Convolutional Neural Networks, which are known for learning the important features of a dataset and are proven to be much more efficient than traditional machine learning method such as K -Nearest-Neighbor or Decision Trees. Results for this research project were compared with similar convolutional neural networks to show that a higher accuracy can be achieved using the proposed method, along with a much simpler design that allows for greater efficiency. Experimental results additionally show that the proposed methodology has serious potential to be implemented in smartphones for continuous authentication. The proposed methodology can achieve an accuracy of 91.2% with an equal error rate of $\mathbf{1 0. 7 \%}$.