Supervised learning methods for biometric authentication on mobile devices

Valerie Ding dingv@stanford.edu

Stephanie Dong sxdong11@stanford.edu

Jonathan Li johnnyli@stanford.edu

We develop fraud detection and user authentication classifiers for mobile keystroke and haptic patterns, achieving 84% accuracy, 90% recall, and 81% precision within one model architecture, and 99% recall and 83% across all models. In addition to proposing these models that outperform existing touch dynamics authentication models, we present a secure, space-efficient, and extensible framework for real-time biometric backlogging comparison.

Complete collaborative repo: https://github.com/jonli123/229-221-joint-project