Transfer Learning for Behavioral Biometrics-based Continuous User Authentication
Sanket Vilas Salunke, Abdelkader H. Ouda, Jonathan Gagné · 2022 International Symposium on Networks, Computers and Communications (ISNCC) · 2022
The cybersecurity industry is developing innovative solutions to avoid cyber-attacks. One such upcoming technology is continuous user authentication. It uses keystrokes and mouse movement behavioral patterns to authenticate the user continuously in the background. This technique uses machine learning to classify users based on the behavioral pattern. It requires a lot of data to find the user’s behavioral pattern and plenty of time is required to gather the data which extends the start of continuously authenticating the new user. In this research, the transfer learning technique was used for a feed-forward neural network model to overcome this issue for new users. Experiments were done using only one behavioral pattern with a set of 5 users to find the difference in accuracy between the model trained with transfer learning and the model trained without any previous learning. The results showed that the model using transfer learning had 9.76% more accuracy than the model trained from scratch. This implies that using transfer learning improves the accuracy with a small amount of data which will help to speed up the onboarding process for new users. This work generates new knowledge which will allow the researchers to implement various machine learning techniques with multiple behavioral patterns, thereby providing the best model performance for transfer learning.