الفهرس | Only 14 pages are availabe for public view |
Abstract User authentication is vital for security in digital services. Organizations now use Multi-Factor Authentication (MFA) to enhance security, using multiple identity checks instead of just relying on Single Factor Authentication (SFA). Keystroke dynamics, analyzing how users type, provides a cost-effective and hardware-independent way to verify users during authentication. This thesis presents two methodologies to improve user authentication using keystroke dynamics. The first methodology introduces an efficient technique using quantile transformation to normalize data distribution, improving accuracy. The main classifier is Histogram Gradient Boosting, achieving high accuracy on the Carnegie Mellon University (CMU) dataset. The second methodology builds on this, proposing an optimized convolutional neural network (CNN) that includes data synthesis and quantile transformation for better feature extraction. The approach uses ensemble learning and achieves exceptional accuracy on the CMU dataset. Combining these approaches creates a comprehensive framework for keystroke dynamics-based user authentication, contributing to strong authentication systems. Practical implications include safeguarding online services and improving security in various domains. |