DoubleType: Authentication Using Relationship Between Typing Behavior on Multiple Devices

Authentication using Keystroke Dynamics (KD), has customarily focused on one device at a time, either desktop or phone. It is imperative that authentication systems adapt to an environment where the users consistently switch between multiple devices. We use the typing behavior of users on different devices, extract the relationship between them and show that these relationships can be used to authenticate users in a multi-device environment when a user switches between devices. We design an authentication system for three scenarios, using the relationship between typing behaviors on a) desktop and phone, b) desktop and tablet, and c) tablet and phone. We find that these are highly separable for individuals with data form 70 users. With Gaussian Naive Bayes (GNB) and Random Forests (RF) classifiers, we found the accuracies for verification to be very high. Using GNB we achieved mean accuracies of 99.15%, 99.23% and 98.72% for relationships between desktop-phone, desktop-tablet and tablet-phone respectively. RF classifiers performed similarly with mean accuracies of 99.31%, 99.33% and 99.12% for relationships between desktop-phone, desktop-tablet and tablet-phone respectively.