Related Experiment Videos
LSTM-based ensemble models for keystroke dynamics authentication: integrating explainable AI for transparency
K Sasikumar1, Sivakumar Nagarajan1
1School of Computer Science and Engineering, Vellore Institute of Technology, Vellore, Tamil Nadu, India.
Frontiers in Artificial Intelligence
|July 31, 2026
Summary
This study introduces a hybrid model for Keystroke Dynamics (KSD) authentication, combining LSTM with ensemble classifiers. The model offers secure, efficient, and interpretable user authentication by analyzing typing patterns.
Area of Science:
- Computer Science
- Cybersecurity
- Machine Learning
Background:
- Traditional authentication methods like passwords and PINs are increasingly insecure.
- Behavioral biometrics, particularly Keystroke Dynamics (KSD), offer a promising alternative for continuous and unobtrusive user authentication.
- KSD analyzes unique typing patterns for identity verification.
Purpose of the Study:
- To develop and evaluate a hybrid model for Keystroke Dynamics (KSD) authentication.
- To enhance the accuracy, efficiency, and interpretability of KSD-based systems.
- To address the growing need for secure and scalable user authentication solutions.
Main Methods:
- A hybrid model combining Long Short-Term Memory (LSTM) networks with an ensemble of Random Forest, XGBoost, and Multilayer Perceptron classifiers was developed.
- Advanced feature engineering techniques were employed to capture typing patterns and user-specific behaviors.
- A soft-voting strategy was utilized within the ensemble classifier.
- SHAP-based explainability analysis was performed to identify key features.
Main Results:
- The proposed hybrid model achieved high authentication performance on the KDA Keystroke Dynamics dataset, with an accuracy of 94.75%.
- The model demonstrated excellent recall (94.89%), precision (94.75%), F1-score (94.76%), and MCC (94.65%).
- Efficient real-time performance was observed, with a throughput of 16,092.93 samples/sec and a detection time of 0.0621 ms/sample.
- SHAP analysis provided interpretability by highlighting influential keystroke features.
Conclusions:
- The hybrid LSTM-ensemble model offers a secure, scalable, interpretable, and computationally efficient solution for real-time Keystroke Dynamics authentication.
- The integration of LSTM for temporal feature extraction and ensemble methods for classification proved effective.
- The study validates the potential of advanced machine learning techniques in enhancing behavioral biometric security.