The Ayushman Bharat Fraud Detection Agent is an intelligent monitoring system designed to detect potential healthcare billing fraud in private hospitals participating in the Ayushman Bharat scheme.

The system analyzes claim data, identifies suspicious patterns such as up-coding, ghost billing, repeat admissions, and abnormal billing behavior, and generates actionable risk scores with explainable reasoning.

This solution combines rule-based logic, anomaly detection models, and hospital-level behavioral risk scoring.

Healthcare insurance fraud causes significant financial leakage. Fraud types include:

• Up-coding (billing higher procedures than performed)
• Ghost billing (fake beneficiaries)
• Repeated admissions manipulation
• Inflated short-stay claims
• Abnormal claim amount deviations

This system provides an automated fraud monitoring engine with explainable risk outputs and investigation support tools.

• CSV-based claim upload
• Column validation
• Duplicate claim detection
• Automatic database insertion

Final Risk Score Formula:

Final Risk Score = 0.45 × Rule Score\

• 0.35 × ML Score\
• 0.20 × Hospital Risk

Components:

• Rule-Based Fraud Detection
  • Short stay high billing\
  • Abnormal procedure-cost deviation\
  • Repeat admissions pattern
• ML-Based Anomaly Detection
  • Isolation Forest (Scikit-learn)\
  • Anomaly scoring using decision function
• Hospital Risk Modeling
  • Percentile-based institutional risk scoring\
  • Aggregated hospital fraud indicators

• Risk-level filtering (High / Medium / Low)
• Claim search by ID
• Hospital drilldown view
• Beneficiary claim timeline
• Monthly fraud trend visualization
• Raw database inspection page

Chronological tracking of claims for a beneficiary allows detection of:

• Repeat admissions within short intervals
• Escalating billing patterns
• Suspicious hospital-specific concentration

CSV Upload
↓
Data Validation
↓
Feature Engineering
↓
Rule Engine
↓
Isolation Forest (Anomaly Detection)
↓
Hospital Risk Aggregation
↓
Weighted Risk Calculation
↓
SQLite Storage
↓
Dashboard Visualization

• Python
• Flask
• SQLite
• Scikit-learn
• Pandas
• NumPy
• Chart.js

ayushman_fraud_detection/
│
├── app.py
├── fraud.db
├── data_generator.py
├── config.py
│
├── database/
│   └── db.py
│
├── services/
│   ├── rule_engine.py
│   ├── anomaly_model.py
│   ├── feature_engineering.py
│   ├── hospital_risk.py
│   ├── risk_engine.py
│   └── data_pipeline.py
│
├── models/
│   └── isolation_model.pkl
│
├── templates/
│   ├── dashboard.html
│   ├── claim.html
│   ├── hospital.html
│   ├── patient.html
│   ├── months.html
│   ├── upload.html
│   └── database.html
│
├── static/
│   └── style.css

1. Install dependencies:

pip install flask pandas scikit-learn numpy

2. Generate sample data:

python data_generator.py

3. Start application:

python app.py

4. Open in browser:

http://127.0.0.1:5000

Unlike basic anomaly detection demos, this system:

• Combines rule-based and ML approaches\
• Incorporates hospital-level risk scoring\
• Provides explainable fraud reasoning\
• Operates as a deployable end-to-end monitoring system

Built with focus on practical fraud detection, explainable AI, and real-world deployable architecture.