language en

Fraud Detection ML-Enabled System — Example Model

Release: 2026-05-03

Modified on: 2026-05-03
License:
http://insertlicenseURIhere.example.org
Visualization:
Visualize with WebVowl
Cite as:
Fraud Detection ML-Enabled System — Example Model.
Provenance of this page
Ontology Specification Draft

Fraud Detection ML-Enabled System — Example Model: Overview back to ToC

This ontology has the following classes and properties.

Named Individuals

Fraud Detection ML-Enabled System — Example Model: Description back to ToC

Example architecture model for a real-time fraud detection system that demonstrates the ML-Enabled Systems extension. Shows how ML components (models, pipelines, feature stores) integrate with traditional software components (payment service, transaction database, notification service).

Cross-reference for Fraud Detection ML-Enabled System — Example Model classes, object properties and data properties back to ToC

This section provides details for each class and property defined by Fraud Detection ML-Enabled System — Example Model.

Named Individuals

ADR-001: Use XGBoost for Fraud Detectionni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#ADR001-ModelChoice

belongs to
decision c

Compliance Officerni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#ComplianceOfficer

Responsible for regulatory compliance of the fraud detection system (PSD2, GDPR).
belongs to
stakeholder c

Drift Detection Response Timeni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#QAS-DriftDetection

belongs to
quality attribute requirement c

Engineered Feature Dataset v2.3ni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#FeatureDataset

Feature-engineered dataset derived from raw transactions — velocity features, aggregates, embeddings.
belongs to
dataset c

Fraud Analytics Team — Data Scientistsni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#FraudDataScientist

Data scientists responsible for developing and improving fraud detection models.
belongs to
stakeholder c

Fraud Case Management UIni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#CaseManagementUI

Web application used by fraud analysts to review flagged transactions.
belongs to
element c

Fraud Detection Model v3.1ni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#FraudDetectionModel

XGBoost gradient-boosted ensemble for real-time transaction fraud scoring. Outputs a fraud probability score (0.0–1.0) for each transaction. Transactions scoring above 0.85 are blocked; 0.60–0.85 are flagged for manual review.
belongs to
m l model c

Fraud Feature Storeni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#FraudFeatureStore

Feast-based feature store providing consistent features for both training and real-time serving. Features include transaction velocity (count/sum over 1h/24h/7d windows), merchant risk scores, and device fingerprint embeddings.
belongs to
feature store c

Fraud Model Fairness Auditni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#QAS-FairnessAudit

belongs to
quality attribute requirement c

Fraud Model Monitorni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#FraudModelMonitor

Evidently AI-based monitoring component tracking: - Prediction distribution (score histogram drift) - Feature drift (PSI on top 20 features) - Model performance (precision/recall on labeled feedback) - Serving latency and error rates
belongs to
monitoring component c

Fraud Model Registryni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#FraudModelRegistry

MLflow Model Registry tracking all fraud model versions, their approval status, and deployment targets.
belongs to
model registry c

Fraud Model Training Pipelineni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#FraudTrainingPipeline

Kubeflow pipeline that trains the fraud detection model: 1. Pull features from feature store 2. Split train/validation/test 3. Train XGBoost ensemble 4. Evaluate on holdout set 5. Register model if metrics pass threshold
belongs to
training pipeline c

Fraud Scoring Integrationni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#integration-payment-scoring

Payment service calls fraud scoring service via gRPC for every transaction. Contract: send TransactionFeatures proto, receive FraudScore proto. Latency budget: 50ms (within the 200ms total authorization window). Fallback: if scoring service is unavailable, approve transaction and flag for async review (fail-open policy agreed with risk team).
belongs to
integrates with c

Fraud Scoring Serviceni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#FraudModelServer

Seldon Core deployment serving the fraud detection model via gRPC. Receives transaction features, returns fraud probability score. Deployed on Kubernetes with autoscaling (2–20 replicas).
belongs to
serving infrastructure c

LSTM Recurrent Neural Networkni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#Option-LSTM

belongs to
option c

MLflow Experiment Trackerni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#FraudExperimentTracker

MLflow instance tracking all fraud model experiments — hyperparameters, metrics, artifacts.
belongs to
experiment tracker c

Model Explainability for PSD2 Complianceni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#Force-Explainability

PSD2 requires that declined transactions can be explained to customers.
belongs to
quality attribute requirement c

No GPU Infrastructureni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#Force-InfraCost

Current Kubernetes cluster has no GPU nodes. Adding GPU support requires 3-month procurement cycle.
belongs to
business constraint c

Notification Serviceni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#NotificationService

Service that sends fraud alerts to customers and compliance team.
belongs to
element c

Payment Processing Serviceni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#PaymentService

Core payment processing service that handles transaction authorization.
belongs to
element c

Payments Team — Software Engineersni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#PaymentsSoftwareEngineer

Software engineers responsible for the payment processing service.
belongs to
stakeholder c

Platform Team — Data Engineersni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#PlatformDataEngineer

Data engineers responsible for ML infrastructure, pipelines, and model serving.
belongs to
stakeholder c

Random Forestni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#Option-RandomForest

belongs to
option c

Real-time Latency Requirementni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#Force-Latency

belongs to
quality attribute requirement c

Sequential Transaction Pattern Detectionni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#Force-SequentialPatterns

Ability to detect fraud patterns across sequences of transactions.
belongs to
quality attribute requirement c

Transaction Databaseni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#TransactionDB

PostgreSQL database storing all payment transactions.
belongs to
element c

Transaction ETL Pipelineni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#TransactionETL

Airflow DAG that extracts transactions from the DB, transforms them, and loads into the feature store.
belongs to
data pipeline c

Transaction Training Dataset v2.3ni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#TransactionDataset

Labeled transaction dataset with fraud/non-fraud labels, covering 18 months of history.
belongs to
dataset c

Which ML algorithm for real-time fraud scoring?ni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#Issue-ModelSelection

belongs to
issue c

XGBoost Gradient Boosted Ensembleni back to ToC or Named Individual ToC

IRI: https://model.example.com/fraud-detection#Option-XGBoost

belongs to
option c

Legend back to ToC

ni: Named Individuals

Acknowledgments back to ToC

The authors would like to thank Silvio Peroni for developing LODE, a Live OWL Documentation Environment, which is used for representing the Cross Referencing Section of this document and Daniel Garijo for developing Widoco, the program used to create the template used in this documentation.