AEGIS PROTOCOL - An Advanced Hybrid Continuous Authentication Approach using Behavioural Biometrics

Abstract

Traditional, static authentication systems leave user sessions vulnerable to post-login threats like session hijacking. This research project, the Aegis Protocol, addresses this gap by developing an advanced continuous authentication (CA) system based on free-text keystroke dynamics. The core of this work is the design and implementation of a novel hybrid deep learning architecture that captures the nuanced, temporal patterns of an individual's typing rhythm to provide persistent, passive security. The methodology involved using the large-scale Buffalo Keystroke Dataset to train the proposed EnhancedHybridModel. A comprehensive feature engineering pipeline was implemented to create a rich 30-dimensional feature vector, which was then fed into a model combining a multi-scale Convolutional Neural Network (CNN), a deep Bidirectional LSTM (BiLSTM), and a self-attention mechanism. The entire system, developed in Python with PyTorch, was realised as a functional proof-of-concept with a demonstration UI and a Flask API. The final model was rigorously evaluated on an unseen test set, achieving a competitive Equal Error Rate (EER) of 22.58%, a security-focused False Acceptance Rate (FAR) of 8.06%, and an Area Under the ROC Curve (AUC) of 0.839. These results, supported by qualitative expert feedback, validate the effectiveness of the proposed architecture. The Aegis Protocol contributes a novel and robust architectural blueprint to the field of behavioural biometrics, demonstrating a powerful approach for developing next-generation continuous authentication systems.