Towards continuous authentication on mobile phones using deep learning models


Volaka H. C. , Alptekin G. , Basar O. E. , Isbilen M., Incel O. D.

16th International Conference on Mobile Systems and Pervasive Computing, MobiSPC 2019, 14th International Conference on Future Networks and Communications, FNC 2019, 9th International Conference on Sustainable Energy Information Technology, SEIT 2019, Halifax, Canada, 19 - 21 August 2019, vol.155, pp.177-184 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Volume: 155
  • Doi Number: 10.1016/j.procs.2019.08.027
  • City: Halifax
  • Country: Canada
  • Page Numbers: pp.177-184
  • Keywords: continuous authentication, mobile sensors, deep learning

Abstract

Smartphones have become essential objects for our daily lives. Besides their original purpose of use, people use these devices as their personal assistants. Additionally, smartphones provide large internal storage which enables users to store their private information, such as personal photos, contact details, call histories, etc. On the other hand, because of their small sizes, these devices could easily get lost or stolen. Therefore, providing the security and privacy of smartphone users against unauthorized access is a significant and crucial area of research. One of the solutions is the use of behavioral biometrics, which tracks and identify users' interaction patterns with the device. In this paper, we investigate the impact of using both touchscreen-based and sensor-based features in an authentication model using deep learning methods. Mainly, we train a three-layer deep network on the combined feature-sets and applied classification for revealing the behavioral characters of users for building an authentication model. We use HMOG dataset that includes data from 100 users over 24 sessions. We train different networks with different combinations of input data, namely only touch-screen data, only sensor data, and their combination. Our results show that we can achieve 88% accuracy and 15% EER values considering binary classification when different types of data are used together. (C) 2019 The Authors. Published by Elsevier B.V.