Key Press Velocities and Depth for User Authentication

Keystroke dynamics is all about finding the differences in how people type on a keyboard. To find these differences, it is necessary to collect data from a user as they type. This research investigates the potential of using additional analog data to improve keystroke dynamics-based user authentication, specifically the velocities of pressing and releasing a key on a keyboard and key press depth, captured with a hall effect keyboard. The study collects data from four subjects, where each subject performed 250 password entries of a fixed password on a hall Effect keyboard. From this data, traditional keystroke metrics such as Hold time (H), Down-Down time (DD) (time between key presses), and Up-Down time (UD) (time between key release and next key press), along with the new analog metrics: Press Velocity Down (VD), Release Velocity Up (VU), and key press Depth (DP), were extracted. Using these collected keystroke metrics, each metric was evaluated with various Machine Learning (ML) models, specifically Random Forest (RF), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost). The results showed that the lowest VD or VU score in accuracy was 84.5%, with its highest accuracy being in the RF model, where VU scored 91% and VD scored 92%. DP scored poorly, with its highest accuracy score being 41.5% and precision up to 51% (RF). Thus, the results indicate that VD and VU can be used as accurate data points for keystroke dynamics, while DP cannot.