Biometric Identification Based on Keystroke Dynamics.

Area of Science:

• Computer Science
• Biometrics
• Machine Learning

Background:

• Keystroke dynamics is a behavioral biometric for user authentication.
• Neural networks offer potential for analyzing complex temporal data like keystrokes.
• Optimizing neural network architecture is crucial for biometric performance.

Purpose of the Study:

• To investigate the influence of neural network architecture and hyperparameters on keystroke dynamics biometric identification.
• To evaluate different neural network layer types and configurations for this task.

Main Methods:

• Utilized a publicly available keystroke dynamics dataset.
• Trained various neural network models including convolutional, recurrent, and dense layers.
• Incorporated pooling and dropout layers in model configurations.
• Compared performance against a state-of-the-art baseline model.

Main Results:

• Performance varied significantly based on network architecture and hyperparameter choices.
• The best-performing model achieved an accuracy of 82% for a 1-of-20 identification task.
• Specific configurations of convolutional, recurrent, and dense layers showed differential impacts.

Conclusions:

• Neural network architecture and hyperparameter tuning are critical factors for effective keystroke dynamics biometrics.
• Further research can optimize these models for improved user identification accuracy.
• The findings provide insights for developing robust behavioral biometric systems.