User Behavioral Biometrics Identification on Mobile Platform using Multimodal Fusion of Keystroke and Swipe Dynamics and Recurrent Neural Network

Concerns of mobile technology security has become an important issue in this era because of the popularity of mobile devices. Private and sensitive data such as photos, contacts, and information for e-banking services are being stored in users' mobile devices. In tradition, single-factor authentication, such as password-based authentication, is employed for identifying who is accessing the device. However, problem happened if the password is stolen. Therefore, numerous researchers have spent efforts in developing and investigating various approaches related to mobile technology security by incorporating of biometrics information. Soft biometrics, such as keystroke and swipe dynamics, and fusion of their features, have been proposed as an alternative solution to traditional biometrics, such as face, iris and voice, in authentication. This paper presents a multi-stream recurrent neural network (RNN) for user identification. Modals to temporal feature, spatial feature and swipe dynamic feature are fused for achieving improved performance with accuracy of 94.26% and F1 score of 93.19%.