Enhancing Keystroke Dynamics Authentication with Ensemble Learning and Data Resampling Techniques

Background: Keystroke dynamics authentication is a behavioral biometric method that verifies user identity by analyzing typing patterns. While traditional machine learning methods (e.g., decision trees, SVM) have shown potential in this field, their performance suffers in real-world scenarios due to data imbalance and limited recognition of certain classes, undermining system security and reliability. Methods: To address these issues, this study combines the Synthetic Minority Over-sampling Technique (SMOTE) with ensemble learning methods to improve classification accuracy. A Django-based platform was developed for standardized keystroke data collection, generating a balanced dataset to evaluate various classifiers. Experiments: Experiments were conducted using both the Django-collected dataset and the CMU benchmark dataset to compare traditional classifiers with SMOTE-enhanced ensemble learning models, such as Random Forest, XGBoost, and Bagging, on metrics like accuracy, recall, G-mean, and F1-score. Conclusions: The results show that SMOTE-enhanced ensemble learning models significantly outperform traditional classifiers, particularly in detecting minority classes. This approach effectively addresses data imbalance, improving model robustness and security, and provides a practical reference for building more reliable keystroke dynamics authentication systems.