A hypo-optimum feature selection strategy for mouse dynamics in continuous identity authentication and monitoring

Mouse dynamics has recently become an interesting new topic in computer security and biometrics due to its non-intrusiveness and convenience. While several pattern recognition methods have been proposed to verify a user based on characteristics of mouse dynamics, they are not applicable to continuous identity authentication and monitoring because most features adopted are statistical-based. This paper compares two hypo-optimum feature selection and evaluation methods to obtain the best combination of features for continuous identity authentication and monitoring. Experiments show that most of the selected feature parameters (12 out of 14) are real time computable which means these features are suitable for online monitoring. Classification results by SVM (Support Vector Machine) show that the performance of feature-selected samples are encouraging with the FAR of 1.86% and FRR of 3.46%, suggesting continuous identity authentication and monitoring with high accuracy is achievable.