Using Deep Learning for Fusion of Eye and Mouse Movement based User Authentication

Yudong Liu, Yusheng Jiang, John Devenere

Abstract: This paper presents a deep learning based user authentication system which aims to identify an individual using data gathered from a mouse and the user's eyes during computer use in a controlled environment. A stacked bidirectional and unidirectional Long Short-Term Memory Recurrent Neural Network (SBV-LSTM-RNN) is introduced to distinguish a legitimate user from impostors. As one of the few attempts of using fusion of mouse and eye movement for user authentication, the proposed system, when adopted on a small dataset, has shown promising improvement compared to a similar system where fusion of eye and mouse modalities and a traditional machine learning method are used.