SecureFin: Video KYC and AML Fraud Detection

SecureFin is an AI-powered academic prototype for automated identity verification (KYC) and financial anomaly detection (AML). It combines computer vision, optical character recognition, facial verification and machine-learning-based anomaly detection to identify suspicious activity.

Key Features

Identity Verification

• OCR-based data extraction from Aadhaar and PAN documents
• Face comparison between an identity document and live capture
• YOLOv8-based person detection
• Multi-person detection for suspicious verification attempts
• Guided liveness verification using prompted actions
• Date-of-birth and name validation
• KYC decisions with confidence scores and reason codes

AML Anomaly Detection

• Isolation Forest model for financial anomaly detection
• Rule-based transaction validation
• Transaction risk scoring
• Real-time decisions: ALLOW, REVIEW or BLOCK
• Model training and evaluation scripts

Security

• Password hashing using bcrypt
• JWT-based authentication
• Fernet encryption for uploaded identity documents
• Environment-based secret configuration
• SQLite-based local data storage
• Uploaded documents, databases and environment files excluded from Git

Evaluation and Analytics

• AML classification report and confusion matrix
• AML performance and distribution visualizations
• KYC status and confidence distributions
• KYC reason-code frequency
• KYC validation-parameter visualization
• JSON-based evaluation metrics

System Workflow

Identity Document and Live Capture
                │
                ▼
      YOLOv8 Person Detection
                │
                ▼
         OCR Data Extraction
                │
                ▼
   Face Matching and Liveness Check
                │
                ▼
         KYC Decision Engine
                │
                ▼
   Isolation Forest + Business Rules
                │
                ▼
         ALLOW / REVIEW / BLOCK

Project Structure

video-kyc-aml-detection/
├── backend/
│   ├── main.py
│   ├── run_server.py
│   ├── aml.py
│   ├── kyc.py
│   ├── liveness.py
│   ├── ocr.py
│   ├── face_match.py
│   ├── yolo_detector.py
│   ├── security.py
│   ├── db.py
│   ├── prepare_data.py
│   ├── train_aml_model.py
│   ├── evaluate_aml_model.py
│   ├── evaluate_kyc_metrics.py
│   ├── aml_model.pkl
│   └── yolov8n.pt
├── frontend/
│   ├── index.html
│   ├── script.js
│   └── styles.css
├── research_graphs/
│   ├── aml_classification_report.txt
│   ├── aml_confusion_matrix.png
│   ├── aml_distribution.png
│   ├── aml_metrics.json
│   ├── aml_performance_graph.png
│   ├── aml_scatter.png
│   ├── kyc_confidence.png
│   ├── kyc_confidence_timeline.png
│   ├── kyc_metrics.json
│   ├── kyc_reason_code_frequency.png
│   ├── kyc_status_distribution.png
│   └── kyc_validation_parameters.png
├── screenshots/
│   ├── aml-block-decision.png
│   ├── aml-review-decision.png
│   ├── kyc_failed_DOB_Mismatch.png
│   ├── kyc_verified_adhar.png
│   └── kyc_verified_pan.png
├── database/                  # Generated locally and excluded from Git
├── frontend_server.py
├── generate_graphs.py
├── run_project.ps1
├── requirements.txt
├── .gitignore
└── README.md

Tech Stack

• Programming Language: Python
• Backend: FastAPI, Uvicorn
• Frontend: HTML, CSS, JavaScript
• Computer Vision: YOLOv8, OpenCV
• OCR: Tesseract OCR, Pytesseract
• Facial Verification: DeepFace
• Machine Learning: Scikit-learn, Isolation Forest
• Database: SQLite
• Security: JWT, bcrypt, Fernet encryption
• Visualization: Matplotlib

Installation

Prerequisites

Install the following software:

• Python 3.10 or later
• Git
• Tesseract OCR

On Windows, Tesseract OCR is commonly installed at:

C:\Program Files\Tesseract-OCR\tesseract.exe

Clone the Repository

git clone https://github.com/kompalwargangotri/video-kyc-aml-detection.git
cd video-kyc-aml-detection

Create a Virtual Environment

python -m venv venv

Activate it on Windows:

venv\Scripts\activate

Activate it on Linux or macOS:

source venv/bin/activate

Install Dependencies

pip install -r requirements.txt

Running the Application

Option 1: PowerShell Launcher

On Windows, run:

.\run_project.ps1

If PowerShell script execution is disabled:

Set-ExecutionPolicy -Scope Process Bypass
.\run_project.ps1

Option 2: Run Manually

Open the first terminal and start the backend:

cd backend
python -m uvicorn main:app --reload --host 127.0.0.1 --port 8000

Open another terminal from the project root and start the frontend:

python frontend_server.py

After starting the application:

• Backend: http://127.0.0.1:8000
• Frontend: http://127.0.0.1:5500
• API Documentation: http://127.0.0.1:8000/docs

Demo Login Accounts

The following accounts are intended only for local demonstration:

Role	Username	Password
Agent	agent01	agent123
Reviewer	reviewer01	review123

Never use these credentials in a publicly deployed application.

Model Evaluation

Run the AML evaluation from the project root:

python backend/evaluate_aml_model.py

Run the KYC evaluation:

python backend/evaluate_kyc_metrics.py

Generate database-driven research graphs:

python generate_graphs.py

Evaluation outputs are stored in the research_graphs/ directory.

Evaluation Visualizations

AML Confusion Matrix

AML Distribution

AML Performance

KYC Confidence

KYC Status Distribution

KYC Reason-Code Frequency

KYC Validation Parameters

Application Screenshots

KYC Verification – Aadhaar

KYC Verification – PAN

KYC Verification – Date of Birth Mismatch

AML Review Decision

AML Block Decision

Data and Privacy

The following files are intentionally excluded from this repository:

• Environment variables and encryption keys
• SQLite database files
• Uploaded identity documents
• Encrypted KYC captures
• Temporary files and logs
• Raw transaction datasets

Use synthetic or properly anonymized data when reproducing the experiments.

Responsible Use

This project is an academic prototype created for learning and demonstration. It should not be used as a production KYC or financial decision-making system without:

• Security auditing
• Regulatory and legal review
• Bias and fairness testing
• Extensive performance validation
• Secure infrastructure and access controls

Future Improvements

• Improve liveness detection against presentation attacks
• Add identity-document forgery detection
• Evaluate the system on larger and more diverse datasets
• Add automated testing and continuous integration
• Containerize the application using Docker
• Deploy the API and frontend securely
• Add role-based access control and audit logging

Author

Gangotri Kompalwar

• GitHub

• LinkedIn

• Portfolio

• Email