Keyboard Data Protection Technique Using GAN in Password-Based User Authentication: Based on C/D Bit Vulnerability.

Area of Science:

• Computer Science
• Cybersecurity
• Machine Learning

Background:

• Password-based authentication remains prevalent due to convenience, despite significant security risks.
• Password exposures are responsible for a substantial majority (82%) of security incidents, highlighting the need for enhanced security measures.
• Existing machine learning-based attacks effectively steal keyboard data, posing a continuous threat to user authentication.

Purpose of the Study:

• To analyze prior research on keyboard data attacks and defense techniques.
• To identify fundamental reasons behind keyboard data attacks and derive effective countermeasures.
• To propose and verify a novel keyboard data protection technique using Generative Adversarial Networks (GAN).

Main Methods:

• Analysis of existing machine learning-based keyboard data theft techniques.
• Development of a novel defense mechanism employing Generative Adversarial Networks (GAN).
• Performance evaluation comparing the proposed GAN-based method against prior machine learning attack strategies.

Main Results:

• Machine learning-based keyboard data attacks achieved a 96.7% success rate in prior research.
• The proposed GAN-based protection technique reduced the attack success rate by approximately 13%.
• Keyboard data classification performance decreased by 29% to 52% on average, and over 50% in maximum performance evaluations.

Conclusions:

• Generative Adversarial Networks (GAN) offer a viable solution for enhancing keyboard data security.
• The proposed method demonstrates significant effectiveness in mitigating sophisticated machine learning-based keyboard data attacks.
• Further research into GAN-based defenses is warranted to combat evolving cyber threats in user authentication.