Advancing Multi-Modal Behavioral Biometric Authentication: A Deep Learning Approach With Synthetic Data Generation
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Behavioral biometrics are recognized as an important component of the next generation of authentication schemes, given their non-invasive characteristics and their ability to provide continuous monitoring as compared to traditional physiological biometrics. Existing behavioral biometric authentication systems, entail significant enrollment costs, privacy issues, and do not have widely available or systematic benchmarks across a multitude of models and a variety of types of behavioral modalities. It is the first manuscript to describe a hybrid GAN-VAE model for generating synthetic data tied to formal differential privacy guarantees (with privacy parameters ε=1.0, and δ=10⁻⁵) and is also the first description of the differential privacy implementation across the nine different behavioral biometric modalities. A comparative analysis was performed of five popular artificial intelligence algorithms, Random Forest, Support Vector Machine, Long Short-Term Memory, Convolutional Neural Network, and hybrid CNN- LSTM, across a cross-modal synthetic data generation methodology that ensures privacy from the mechanism and that retains inter- modal relationships. The Authors findings show that generated data can reduce the enrollment threshold by up to 75% and the selected models can achieve >90% accuracy in pairwise authentication measurement. The hybrid CNN-LSTM model had the best overall performance having an average accuracy of 96.04% among all modalities, followed by LSTM (94.32%) and CNN (92.51%). A hybrid approach achieved the optimal performance in keystroke dynamics with an accuracy of 99.12%. Six recent benchmark datasets published between 2023 and 2025 are used for external validation, where the system outperforms the state of the art baselines by a mean of +5.47% on all modalities. This study presents new methods for maintaining cross-modal consistency, thorough demographic bias analyses, and application strategies that have been tested and refined through active case studies. These findings carry important implications for the development of usable, large-scale, and privacy-preserving behavioral biometric authentication systems with minimal associated data collection overhead.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.002 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.002 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it