Touchscreen-based Smartphone Continuous Authentication System (SCAS) using Deep Neural Network

Abstract

Due to the huge increasing usage of mobile devices and applications, basic authentication is not a secure choice for a long user session. Therefore, it needs a second layer of protection developed as a seamless and non-intervention security procedure, which is the continuous authentication. A framework analyses behavioral biometrics dataset was designed based on the touch-screen and a user continuous authentication was applied depending on pre-specified acceptance ratio to the portions of the stroke. In this framework, an approach based on a deep learning technique was adopted to train a model able to classify these behavioral data and achieve high accuracy. By comparing with the traditional use of machine learning algorithms that required more processing actions, which it is considered costly from the perspective of execution and the complexity of processing. In general, machine-learning algorithms need to extract statistical features from the raw data before they can work on it directly. The experiences of using machine learning classification algorithms have proven unreliable results when they are using on one behavioral biometrics modality, comparison with our framework that are based on a deep learning technique. The approach designed to apply a deep neural network on behavioral biometric data such as X, Y coordinates; area covered and phone orientation …etc. These features are captured directly from the touchscreen sensor of the smartphone during the user’s sessions, to train system able to classify new input data with high accuracy if it is coming from a genuine or fraudulent user, which already was achieved and obtained test accuracy 94.2% with equal error rate (EER 3.46%).