Ensemble Learning to Enhance Continuous User Authentication For Real World Environments
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
Modern digital applications/systems need robust cybersecurity solutions. Traditional authentication methods like passwords, fingerprints, authorization cards, etc. authenticate the user at the beginning of the session but there is no validation during the session, which makes the system vulnerable. Continuous authentication is the solution to this challenge. In continuous authentication, keystroke data is used to extract the behavior patterns of the user. The data are then applied to train the machine learning (ML) classification algorithms to identify the unique behavioral patterns of each user and classify them accordingly. Thus, the performance of the ML classification algorithm is key in continuous user authentication, and it requires diverse and comprehensive data to be effective in the production environment. In many cases, the ML algorithm is trained on the datasets collected in a controlled lab environment and the model fails or does not perform as expected in the production environment. For example, China's facial recognition system recognized the face on a bus ad as a jaywalker because the model was not trained on real-world data. To overcome this problem, this study uses the real-world data of 48 of a financial organization's employees to compare the prediction accuracy and prediction delay (the time required to make predictions) of advanced ML algorithms, including Light GBM, XGboost, TabNet, Neural Network, and 1D CNN. Among all the individual models, LightGBM performed best with an accuracy of 23.58% and a delay of 34.4 sec. However, some ML models were better at predicting particular sets of users than others, hence ensemble learning was used to combine the prediction ability of all the models, which increased cumulative accuracy to 24.03% with a delay of 43.51sec. These results suggest that the boosting algorithm is effective at classifying users. Additionally, the prediction performance can be improved using ensemble learning techniques. Moreover, high-end infrastructure should be used to reduce the prediction delay of ML algorithms.
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 | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.002 |
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