HealthFraudML — Open-Source Healthcare Fraud Detection Framework

Healthcare fraud costs the U.S. over $100 billion annually. HealthFraudML is a modular Python framework that brings machine learning to the front lines of fraud detection — from claims-level ML models to patient-facing billing audits with auto-generated dispute letters.

What Can It Do?

from healthfraudml import BillingAuditor

# Audit a patient bill for upcoding, unbundling, and price gouging
auditor = BillingAuditor(provider_name="Example Health System")
report = auditor.audit_bill([
    {"cpt_code": "99285", "amount": 6672.00, "description": "ED Visit Level 5"},
    {"cpt_code": "56420", "amount": 709.00, "description": "Bartholin Cyst I&D"},
])

print(f"Risk Level: {report['risk_level']}")           # "High"
print(f"Potential Savings: ${report['suggested_savings']:.2f}")  # "$5,472.00"
print(report['dispute_letter'])                         # Ready-to-send dispute letter

Try it now → Open in Google Colab

Key Features

Supported Fraud Types

Installation

pip install healthfraudml

Or install from source:

git clone https://github.com/bharath309/healthfraudml.git
cd healthfraudml
pip install -e .

Quick Start — Claims-Level Fraud Detection

from healthfraudml import FraudDetector
from healthfraudml.models import HybridEnsemble
from healthfraudml.preprocessing import ClaimsPreprocessor

# Load and preprocess claims data
preprocessor = ClaimsPreprocessor()
X_train, X_test, y_train, y_test = preprocessor.load_and_split("claims_data.csv")

# Train hybrid ensemble model
detector = FraudDetector(model=HybridEnsemble())
detector.fit(X_train, y_train)

# Detect fraud with explainable outputs
results = detector.predict(X_test, explain=True)

for case in results.flagged:
    print(f"Claim {case.id} | Score: {case.score:.3f} | {case.explanation}")

Quick Start — Patient Billing Audit with RAG

Note: the RAG and LLM features need extra dependencies. Install them with pip install "healthfraudml[rag]" (adds ChromaDB, Google GenAI, and pypdf).

from healthfraudml import CPTDatabase, RAGBillAuditor

# Initialize vector database with CPT pricing rules
db = CPTDatabase()

# RAG auditor: retrieves rules from ChromaDB → prompts Gemini → generates audit
rag = RAGBillAuditor(db=db)
report = rag.audit_bill([
    {"cpt_code": "99285", "amount": 6500.00, "description": "ED Level 5"},
    {"cpt_code": "",      "amount": 900.00,  "description": "Bartholin cyst drainage"},
], provider_name="Example Health System")

# Semantic search resolves "Bartholin cyst drainage" → CPT 56420 automatically
print(f"Savings: ${report['suggested_savings']:.2f}")
print(report['dispute_letter'])

Organizational Readiness Assessment

from healthfraudml.readiness import ReadinessAssessment

assessment = ReadinessAssessment()
report = assessment.evaluate(
    institution_size="medium",
    annual_claims_volume=120000,
    existing_fraud_detection="manual",
    it_staff_count=5,
    annual_fraud_budget=45000
)

print(f"Readiness Score: {report.readiness_score}/100")
print(report.recommendations)
print(report.implementation_roadmap)

Architecture

healthfraudml/
├── detector.py              # Main FraudDetector API
├── models/
│   ├── supervised/          # Neural Net, SVM, Random Forest, Gradient Boosting, Bayesian
│   ├── unsupervised/        # K-Means, Outlier Detection, Artificial Immune Systems
│   └── hybrid/              # AdaBoost Ensemble, Stacked Models
├── preprocessing/           # Claims pipelines, feature engineering, HIPAA privacy
├── evaluation/              # Metrics, benchmarking, SHAP/LIME explainability
├── readiness/               # TAM-based organizational assessment
├── auditor/                 # Patient billing auditor
│   ├── billing_auditor.py   # Rule-based CPT audit engine
│   ├── llm_integration.py   # Gemini LLM parser + RAG auditor
│   └── db.py                # ChromaDB vector store
├── fraud_types/             # Upcoding, phantom billing, unbundling, etc.
└── examples/                # Jupyter notebooks and demo scripts

Research Foundation

Built from doctoral research at the University of the Cumberlands examining ML adoption for healthcare fraud detection across U.S. institutions. Grounded in Technology Acceptance Model (TAM), Fraud Triangle Theory, and Diffusion of Innovations Theory.

Citation:

@phdthesis{bahudhoddi2025financial,
  title={Financial Fraud Detection in Healthcare Settings: Comparative Analysis
         through Machine Learning to Identify Problems Relating to Fraudulent
         Activities in Hospitals},
  author={Bahudhoddi, Bharath Kumar},
  year={2025},
  school={University of the Cumberlands}
}

Documentation

Full docs are powered by MkDocs with Material theme:

pip install mkdocs mkdocs-material
mkdocs serve
# Open http://127.0.0.1:8000/

Contributing

Contributions are welcome! See CONTRIBUTING.md for guidelines and good first issues.

License

MIT License — see LICENSE for details.

Contact

Bharath Kumar Bahudhoddi, Ph.D.

• Email: bharath.p90@gmail.com