Cost-Effective Kernel Ridge Regression Implementation for Keystroke-Based Active Authentication System.

Area of Science:

• Machine Learning
• Biometrics
• Computer Security

Background:

• Large-scale active authentication systems require efficient algorithms.
• Traditional methods like Support Vector Machines (SVM) with Gaussian Radial Basis Function (RBF) kernels can be computationally expensive for large datasets.

Purpose of the Study:

• To introduce a fast Kernel Ridge Regression (KRR) learning algorithm for large-scale active authentication.
• To implement a truncated Gaussian Radial Basis Function (TRBF) kernel for improved cost-performance.
• To evaluate the computational advantages and error rate performance compared to traditional SVM.

Main Methods:

• Adoption of a fast Kernel Ridge Regression (KRR) learning algorithm.
• Implementation of a truncated Gaussian Radial Basis Function (TRBF) kernel.
• Comparative analysis against Support Vector Machine (SVM) with Gaussian-RBF kernel.

Main Results:

• The fast-KRR with TRBF kernel achieved an Equal Error Rate (EER) of 1.39% with significantly reduced training time.
• SVM with Gaussian-RBF kernel showed a comparable EER of 1.41% but with higher training time.
• The developed authentication system demonstrates cost-effectiveness.

Conclusions:

• The fast-KRR algorithm combined with the TRBF kernel provides a computationally advantageous and cost-effective solution for large-scale active authentication.
• This approach maintains high performance in terms of error rates while reducing computational costs.