Derived Mouse Dynamics Dataset: A Feature-Enhanced Dataset for Behavioral Biometrics and Continuous Authentication

Résumé fourni par la source

The Derived_Mouse_Dynamics_dataset is a feature-enhanced dataset developed for research and intends to provide a richer feature representation of mouse behaviour for application in user profiling, continuous authentication, behavioural analysis, anomaly detection and machine learning approaches. The hypothesis is that individual users exhibit distinctive and recurring mouse-interaction characteristics across sessions and that a combination of behavioural features may provide a stronger and more reliable representation in authentication. A python script was used to process the original Balabit mouse event-logs. Behavioural features were computed from the original timestamps, coordinates and mouse-event information. The resulting dataset is organised into training_files folder, and test_files folder. Each folder comprises of data for 10 users, identified by their respective user IDs (7, 9, 12, 15, 16, 20, 21, 23, 29, 35). Each user folder contains multiple session folders (identified with IDs like session_0041905381) , where each session represents a separate mouse-interaction session. Every session folder contains three CSV files containing the extracted behavioural features (such as session_0041905381_dwell.csv, session_0041905381_dynamics.csv and session_0041905381_flight.csv). The extracted features are classified into three categories : Timing-Based Features - Flight Time, Dwell Time, Left Click Dwell Time, Right Click Dwell Time and Delta Time, computed from the timestamps and mouse button events to capture click duration and time intervals between consecutive mouse events; Spatial Features - X Displacement and Y Displacement, which represent horizontal and vertical cursor movement; and Kinetic Features - X Velocity, Y Velocity, X Acceleration, Y Acceleration, X Jerk and Y Jerk, computed from the coordinate data and time intervals to describe movement speed, changes in speed and abruptness or smoothness of cursor movement. The data shows observable differences in the mouse-interaction patterns between users as well as variation across sessions of the same user. Timing features reveal differences in clicking rhythm and time-based variability while spatial trajectories show difference in movement density and patterns. These observations indicate mouse interaction contains potential distinctive characteristics that can effectively contribute to user profiling. The dataset can be used for continuous authentication, user identification and verification, behavioural profiling, anomaly detection, feature analysis and machine learning experiments. Researchers can perform intra-user analysis by comparing sessions of the same users and inter-user analysis by comparing the behavioural patterns between users. The dataset should be interpreted as resource for investigating behavioural characteristic and their potential for application of user profiling and behavioural authentication; individual features should not be sufficient on its own.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.