Outlier detection for keystroke biometric user authentication.

Area of Science:

• Computer Science
• Cybersecurity
• Biometrics

Background:

• User authentication is critical for information security, facing challenges with identity fraud and data breaches.
• Keystroke dynamics research traditionally relies on imposter datasets, which are difficult to obtain in real-world scenarios.

Purpose of the Study:

• To present a novel approach for keystroke dynamics-based authentication using unsupervised outlier detection.
• To eliminate the necessity of imposter samples in authentication systems.

Main Methods:

• Utilized unsupervised outlier detection techniques, specifically the histogram-based outlier score (HBOS).
• Compared HBOS against 15 alternative outlier detection methods.
• Validated the approach on the Carnegie Mellon University (CMU) keystroke biometrics dataset.

Main Results:

• HBOS demonstrated superior performance compared to 15 other outlier detection methods.
• Achieved an equal error rate (EER) of 5.97%.
• Obtained an area under the ROC curve (AUC) of 97.79% and an accuracy (ACC) of 89.23%.

Conclusions:

• The proposed HBOS method offers a significant advancement in keystroke dynamics authentication by removing the need for imposter data.
• This approach enhances practical applicability and addresses real-world challenges in simulating fraudulent keystrokes.
• The method provides a reliable and efficient solution with improved accuracy and robustness for user authentication.