Family
Isolation Forest + LSTM Autoencoder + Z-Score
in-developmentApache 2.0tabularboth
Anomaly Detector.
Real-time transaction anomaly / fraud detection — Isolation Forest + LSTM Autoencoder + Z-Score
Foundation · Isolation Forest + LSTM Autoencoder + Z-Score
What it does
The honest description.
Real-time transaction anomaly / fraud detection — Isolation Forest + LSTM Autoencoder + Z-Score
Architecture
How it works.
Per-user transaction sequences (60+ days), global transaction features (amount, merchant, category, time-of-day), historical Z-score stats per category. Anomaly score 0-1 (higher = more anomalous) from ensemble of Isolation Forest, LSTM Autoencoder, and Z-Score. Target < 5% false positive rate.
Training data & drift
What it learned from, and how we watch it.
Source
Inputs: Per-user transaction sequences (60+ days), global transaction features (amount, merchant, category, time-of-day), historical Z-score stats per category.
Drift baseline · current quarter
Drift baseline not yet documented for this model.
What it's NOT
- Unsupervised — flags unusual patterns without labeled ground-truth.
- Distinct from fraud_detection which is supervised.
- Self-healing via TabularStrategy (PSI/KS drift) — monitoring enabled per Phase 2 D-DRIFT-2.
Who calls this model
Across the house and the API.
Internal routes
No public routes documented yet.
Featured in
/features/transactions-insightsVersion history · immutable once published
Every release, traceable.
Data Mukodi holds about you is subject to DPDP §12 erasure (use /consent in the app), and any concern about this model can be raised via [email protected] under DPDP §13 grievance, or via /security.