On the Best Sensor for Keystrokes Inference Attack on Android

Abstract One of the most recently exposed security threats on smartphone platforms is the potential use of motion sensors to infer user keystrokes. Exploited as side channels, few researchers have demonstrated the ability of built-in accelerometers and gyroscopes in particular, to reveal information related to user input, though the practicality of such an attack remains an open question. This paper takes further steps along the path of exploring the aspects of the new threat, addressing the question of which available sensors can perform best in the context of the inference attack. We design and implement a benchmark experiment, against which the performances of several commodity smartphone-sensors are compared, in terms of inference accuracy. All available Android motion sensors are considered through different settings provided by the OS, and we add the option of fusing several sensors input into a single dataset, to examine the amount/lack of improvement in the attack accuracy. Our results indicate an outstanding performance of the gyroscope sensor, and the potential improvement obtained out of sensors data fusion. On the other hand, it seems that sensors with magnetometer component or the accelerometer alone have less benefit in the adverted attack.