Performance evaluation of implicit smartphones authentication via sensor-behavior analysis

The pervasiveness of mobile devices not only facilitates people's daily life with a wide variety of services, but also brings users risks of private information leakage (e.g., photos, contact lists, bank accounts), which emphasizes the demand for reliable, feasible and user-friendly authentication mechanisms on mobile devices. In this paper, we develop an authentication mechanism using motion sensors (accelerometer and gyroscope) embedded in smartphones. Our proposed mechanism performs authentication continuously and implicitly by monitoring the user daily activities. We extract time-, frequency- and wavelet-domain features from motion-sensor data, and conduct empirical feature analysis to investigate the optimal combination of features, to acquire a fine-grained characterization of users' movement patterns. To make a systematic performance evaluation, we have established a dataset containing 27,681 samples, including five kinds of actions and five different smartphone placements. In the evaluation procedure, four kinds of contexts are considered (nothing-aware context, action-aware context, placement-aware context and full-information-aware context), and ten one-class detectors are implemented. The best accuracy (represented as EER) for the four conditions achieves 28.22%, 2.21%, 5.50% and 3.28%, respectively, indicating our proposed approach is feasible and applicable in some real scenarios. Moreover, the performance analyses for sensor combinations and feature combinations are also conducted.