Context-Aware Implicit Authentication of Smartphone Users Based on Multi-Sensor Behavior

Implicit authentication is a new research direction to enhance the privacy protection of smartphones. However, implicit authentication has low robustness due to its vulnerability to environment. Motivated by this, this paper proposes a context-aware implicit authentication, which is a scheme to improve the robustness of authentication by introducing context awareness module. In the scheme, multi-sensor data (including accelerometer, gyroscope, magnetometer, timestamp, pressure, touch size) is first captured in a fine-grained manner to characterize one's touch action. Then, based on various sensors data, gesture features and touch features are extracted by employing both statistical method and distance measurement method. Particularly, one's body posture when touch action happens can be used as context. In each context, we present a weighted sum fusion rule to fuse the results of different features obtained by One-Class SVM (OC-SVM). We have collected 10000+ sensor data from 87 participants for experimental evaluation. The results show that the authentication in an unrestricted environment can achieve a best equal error rate (EER) of 0.0071%, which is more than one percent lower than the non-context-aware authentication. The proposed method can effectively improve the reliability and practicability of implicit authentication.