Continuous Authentication of Smartphone Users using Machine Learning

Current smartphone security techniques offer less reliability. As an instance, personal identification numbers can be easily guessed or hacked, fingerprint scan requires a hardware to operate, and face recognition can be affected by light, other people in the background, or different poses by the users. In addition, they are befitting for one-time security, therefore commonly used at the time of login to verify users. However, what if there's a change of user while accessing the smartphone, and the phone is accessed by an intruder after login. To deal with this issue, continuous authentication is applied which regularly and unnoticeably will address the challenges of verifying users via behavioral features, such as keystroke, hand, and orientation activities. The goal of this research project is to design and implement a behavior-based security method and detect intrusion using machine learning. Hand-movement, grasp, and orientation are three behavioral features that can be effectively used to continuously authenticate users. In-built inertial sensors including accelerometer, gyroscope, magnetometer, and orientation are used to unnoticeably represent sensitive micro-movements of hand and orientation pattern when a user accesses the smartphone screen. The researchers in this project investigated large datasets of different smartphone users with different interaction sessions. To detect the behavior of smartphone users, the research applies various supervised machine learning algorithms on the dataset of smartphone users. The experimentation results indicate that the presented approach is promising and can be effectually implemented for continuous authentication of smartphone users.