Supervised classification methods applied to keystroke dynamics through mobile devices

Keystroke dynamics biometrics through computers are based in the time that users need to press and hold keys and often present too small amount of information. This limitation is eliminated in the environment of mobile devices due to a variety of sensors (accelerometers, gyroscopes, pressure and finger size) can be used to acquire useful information from users. These data have been acquired within the scenario of typing a 4-digit PIN in order to analyze the possibilites of reinforcing the security of mobile devices. A database with keystroke dynamics patterns has been analysed. Data has been acquired in a constrained environment, where users must hold the phone in a fixed position, and other with the data taken in unconstrained conditions. Features as pressure, finger size, times, linear an angular acceleration are extracted and processed. Supervised classification methods are widely used in different kind of biometrics. A discussion about their use in keystroke biometrics is presented. A preprocessing of the acquired data is performed using Linear Discriminant Analysis (LDA) and a reduction of the amount of information applying Principal Components Analysis (PCA). This preprocessing enhances considerably the results obtained in classification. We conclude claiming that biometric systems through keystroke dynamics with 4-digit PIN are promising.