DM-FAD: Federated Dual-Modal Anomaly Detection for Healthcare Insurance Fraud Analytics

1. Overview

DM-FAD is a federated learning-based fraud detection system integrating structured and unstructured data modalities. It ensures privacy preservation by decentralizing model training while achieving high fraud detection accuracy.

2. System Components

Component	Function	Technology Used
Structured Pipeline	Processes claims data	Autoencoder
Text Pipeline	Processes OCR documents	BERT
Fusion Layer	Combines outputs	Weighted scoring
Federated Layer	Aggregates models	FedAvg
Security Layer	Ensures privacy	DP + Encryption

3. Requirements

Library	Purpose	Version
Numpy	Numerical operations	1.24.4
Pandas	Data processing	2.0.3
scikit-learn	ML models	1.3.0
Matplotlib	Visualization	3.7.2
Torch	Deep learning	latest
Transformers	BERT models	latest
Pytesseract	OCR	latest
opencv-python	Image processing	latest

4. requirements.txt

numpy==1.24.4 
pandas==2.0.3 
scikit-learn==1.3.0 
matplotlib==3.7.2 
torch 
transformers 
pytesseract 
opencv-python 

5. Folder Structure

DM-FAD-PROJECT/ 
├── FRAUD_ANALYTICS.ipynb 
├── requirements.txt 
├── README.md 
├── DATA/ 
│   ├── claims.csv 
│   └── documents/ 
└── outputs/ 

6. Execution Steps

## 6.1. Install dependencies
## 6.2. Place dataset
## 6.3. Run Jupyter
## 6.4. Execute notebook

7. Outputs

Output	Description
Fraud Score	Probability of fraud
Anomaly Score	Reconstruction error
Embeddings	Text semantic vectors
Graphs	Loss visualization