Biometric Authentication by Keystroke Dynamics for Remote Evaluation with One-Class Classification

One-Class SVM is an unsupervised algorithm that learns a decision function from only one class for novelty detection: classifying new data as similar (inlier) or different (outlier) to the training set. In this article, we have applied the One-Class SVM to Keystroke Dynamics pattern recognition for user authentication in a remote evaluation system at Laval University. Since all of their students have a short and unique identifier at Laval University, this particular static text is used as the Keystroke Dynamics input for a user to build our own dataset. Then, we were able to identify weaknesses of such a system by evaluating the recognition accuracy depending on the number of signatures and as a function of their number of characters. Finally, we were able to show some correlations between the dispersion and mode of distributions of features characterizing the signatures and the recognition rate.