For a simple test of our system, we gathered several hours worth of keystroke data from various users. We used the data from one user to train the model, and then ran predictions both on that user's typing and on the typing of another user. As illustrated by figure 4, all three graphs show significantly higher probabilities when the typist is the same person as trained the model (on the left) than when another person types (on the right).
Figure 4: Predicting Identity Based on a Learned Model