Passive User Identification and Authentication with Smartphone Sensor Data

A unique digital identity, user ID, is essential for everyday online activities in the Internet era. These user IDs represent a user in a digital environment using stored credentials on a system called authentication system. It is possible to capture unique patterns of user movements from smartphone sensor data. This paper presents a framework for passive user identification and authentication using onboard sensors of an Android smartphone. Using this framework, we propose a data preprocessing scheme that uses the absolute difference of consecutive repeated measurements of 7 onboard sensors. We developed 5 models for user identification and authentication using various machine learning and deep learning methods. The experimental results and performance assessment conclude that the Sequential Neural Network (SNN) model performs best with 98.24% accuracy for authenticating users (binary classification) and 84.41% accuracy during multi-class classification for user identification. Furthermore, all the models developed for this research, namely MLP, SNN, CNN, SVM, and Bi-LSTM, provide 100% precision during binary classification for passive user authentication. Our models were trained on 556,746 sensor data samples, each having 20 features. These features include the proximity sensor data, light sensor data, triaxial measurements from accelerometers, gravity sensors, gyroscopes, magnetometers, and rotational vector sensors. We study the possible influence of absolute difference samples for user identification and authentication.