Continuous authentication with feature-level fusion of touch gestures and keystroke dynamics to solve security and usability issues

Abstract

Behavioral Biometrics (BB) Continuous Authentication (CA) systems monitor user behavior and continuously re-authenticate user identity alongside the initial login process. Most studies use single behavioral modality systems to authenticate users. However, the behaviors of genuine users may change, and systems fail when significant changes occur. This results in either usability or security issues. In the literature, the fusion of biometrics is used to solve this problem and achieves improved results. This paper presents our research on the design and evaluation of new approaches to CA using fusion of touch gestures and keystroke dynamics. To collect the biometric data from mobile device users we have developed the BioGames App which follows an innovative approach based on the gamification paradigm. We examine each modality separately and investigate if we can improve the performance results with a feature-level fusion. For this reason, a new appropriate feature set is developed that combines touch gestures and keystroke dynamics. We used the Multi-Layer Perceptron (MLP) and Long Short-Term Memory (LSTM) and compared their performance. We have shown that feature-level fusion of touch gestures and keystroke dynamics improves the performance of systems and solves security and usability issues. We found that the MLP is superior to LSTM in this context. The MLP achieved Accuracy 98.3% (increased 21.1%), EER 1% (error reduction by 23.7%), TAR 99.4% (increased 46%), TRR 97.4% (increased 10%), FAR 2.6% (reduced by 10.5%), and FRR 0.6% (reduced by 46%).