Deep Learning-Driven User Legitimacy Prediction Using Keystroke and Mouse Behavioural Dynamics

Continuous authentication, based on behavioral dynamics such as keystroke and mouse movement patterns, has emerged as a promising solution to detect insider threats and ensure real-time security in sensitive systems. This paper presents a comprehensive comparison of traditional machine learning methods, including Random Forest (RF), K-Nearest Neighbors (KNN), Logistic Regression (LR), and XGBoost, against deep learning models like Deep Feedforward Neural Networks (DFFNN) and Long Short-Term Memory (LSTM) networks. Both keystroke and mouse dynamics datasets were used to evaluate model performance. The LSTM model consistently outperformed all other approaches, achieving the highest accuracy and precision, particularly on the keystroke dynamics dataset with an accuracy of 88%. Additionally, we highlight the decision time of each model, crucial for real-time authentication systems. Our findings emphasize the advantage of deep learning models in handling sequential data, while traditional models showed slower response times and lower accuracy. The paper concludes with a discussion on the potential for combining both keystroke and mouse dynamics for enhanced continuous authentication.