Efficient Convolutional Neural Network-Based Keystroke Dynamics for Boosting User Authentication.

Area of Science:

• Computer Science
• Cybersecurity
• Machine Learning

Background:

• User authentication is critical for online service security.
• Multi-factor authentication enhances security but can be complex.
• Keystroke dynamics offers a seamless authentication method by analyzing typing patterns.

Purpose of the Study:

• To propose an optimized convolutional neural network (CNN) for improved feature extraction in keystroke dynamics.
• To enhance the accuracy and efficiency of user authentication systems.
• To leverage data synthesis and quantile transformation for maximizing authentication results.

Main Methods:

• Utilized an optimized convolutional neural network (CNN) architecture.
• Employed data synthesization and quantile transformation for feature enhancement.
• Applied an ensemble learning technique for model training and testing.
• Evaluated the method on a publicly available Carnegie Mellon University (CMU) dataset.

Main Results:

• Achieved an average accuracy of 99.95%.
• Reached an average equal error rate (EER) of 0.65%.
• Obtained an average area under the curve (AUC) of 99.99%.
• Outperformed recent advancements on the CMU dataset.

Conclusions:

• The proposed optimized CNN with ensemble learning demonstrates superior performance for keystroke dynamics-based authentication.
• The method provides a highly accurate and efficient solution for safeguarding online services.
• Data synthesization and quantile transformation significantly improve feature extraction for behavioral biometrics.