Privacy-preserving and robust mouse dynamics authentication using hybrid transformer-CNN and federated learning
Abstract
Traditional methods like passwords and PINs are increasingly vulnerable, making continuous authentication essential. Behavioral biometrics such as mouse dynamics provide a non-intrusive way to verify users through their unique interaction patterns. This study proposes a secure and privacy-preserving framework for mouse dynamics authentication using a Hybrid Transformer-CNN model, where the Transformer captures global behavioral dependencies and the Convolutional Neural Network (CNN) extracts local movement patterns. The framework includes a comprehensive preprocessing pipeline with feature engineering, SMOTE-based class balancing, and stratified cross-validation. To enhance security and privacy, the system integrates adversarial robustness evaluation and federated learning with differential privacy, enabling decentralized training without sharing raw user data. The model is further evaluated under non-IID data distributions and open-set authentication scenarios with per-user thresholds to reflect realistic deployment conditions. Experimental results demonstrate high performance, achieving 99.51% accuracy, 99.53% precision, low FPR (0.000256), and EER (0.000068). The model maintains robustness under adversarial attacks and achieves efficient real-time detection. Overall, the proposed framework provides a reliable and privacy-aware solution for continuous authentication in security-sensitive applications.
Read the paper
Related studies
- Active Learning during Federated Learning for Object Detection
- Fairness-aware federated learning
- Decentralized Federated Learning
- Building a Personally Identifiable Information Recognizer in a Privacy Preserved Manner Using Automated Annotation and Federated Learning
- Preservation of Patient Level Privacy: Federated Classification and Calibration Models
- Efficient explainable federated learning
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.