AttendanceGuard Ensemble Model
Hybrid Isolation Forest + LSTM Autoencoder for attendance fraud detection.
Intended Use
- Flag suspicious employee-day profiles from turnstile/badge logs
- Support HR and physical security audit workflows
- Not for automated disciplinary action without human review
Model Components
| File | Description |
|---|---|
isolation_forest.joblib |
Tabular outlier detector |
scaler.joblib |
Feature standardizer |
lstm_ae.pt |
Sequence autoencoder weights |
seq_scaler.joblib |
Sequence feature scaler |
config.json |
Thresholds and ensemble weights |
Limitations
- Trained on synthetic data only
- Does not replace CCTV verification or legal HR process
- Performance varies by facility layout and shift policies
Training
python scripts/generate_samples.py
python scripts/train_models.py