A Comparison of Machine Learning Algorithms in Keystroke Dynamics

Usernames and passwords have become weaker as the methods used to bypass this system have advanced. Today, secondary biometric authentication systems can be used to mitigate these vulnerabilities. Biometrics are unique traits that distinguish every individual from each other. One biometric technique is keystroke dynamics which is an authentication method based on a user's typing rhythm on a keyboard. These rhythm patterns are based on digraphs or the timing between two successive key presses. Our goal was to record the keystroke patterns of each user to determine if the users can be authenticated accurately by this method using machine learning. We recorded the latencies between keystrokes and taught machine learning algorithms the datasets to determine if the pattern can be learnable by the machine. We tested four different machine learning algorithms: Decision Trees, Random Forest, Support Vector Machines, and Neural Networks, to determine which is most effective on accuracy. We also tested two text sizes to compare each algorithm's prediction rate based on input size.