Sensor-based continuous user authentication on smartphone through machine learning

With the rapid development of hardware and software technology, many end-users have often stored their data on smartphones through several in-built sensors. Thus, the security of stored data on smartphones has become a significant concern. This has fueled the importance of entry-point authentication methods on smartphones. Many entry-point authentication methods have failed to offer security due to insider or side-channel attacks. This article introduces a sensor-based continuous user authentication approach on the smartphone through multi-modal behavioral biometrics and a machine learning model to tackle the aforementioned issues. The proposed approach captures the touch and motion-based behavioral biometrics through the touchscreen and inertial sensors of the device. Then, the proposed approach extracts several features from captured behavioral data and selects the best set of features through a filter-based feature selection technique. Further, we implement a nonlinear support vector machine with optimized hyperparameters for training and predicting the features to generate the scores. We apply score-level fusion on generated scores of several sensors to compute the final score for identification of the genuine user. In this article, we systematically evaluate the proposed approach with the most commonly used behavioral activities of the smartphone user in our daily life. The experiment results on all behavioral activities show that the proposed approach obtained the best authentication score compared to the other machine learning models, and state-of-the-art methods. Finally, we conclude our article by addressing the limitations of the proposed approach and practical research issues for future exploration.