🛡️ VeriSafe – Hybrid AI Fraud Detection System

Overview

VeriSafe is a hybrid fraud detection system designed to identify suspicious financial transactions using a combination of machine learning models and rule-based validation techniques. The system analyzes transactional data to detect anomalous patterns and classify transactions as fraudulent or legitimate.

The project demonstrates how predictive models can be integrated into a modular backend architecture to support real-time fraud detection workflows.

---

Key Features

• Hybrid fraud detection pipeline combining Machine Learning + Rule-Based Validation
• Implementation of Random Forest and XGBoost classifiers
• Feature engineering and anomaly detection on transaction datasets
• Flask ML microservice for fraud prediction
• Spring Boot backend for API communication
• Modular architecture enabling scalable fraud detection systems

---

System Architecture

---

System Workflow

1. Transaction data is submitted through the frontend.
2. The backend processes the request and sends transaction data to the ML microservice.
3. The machine learning model analyzes the transaction features.
4. Fraud probability is calculated.
5. The result is returned to the backend and displayed to the user.

---

Machine Learning Pipeline

Data Preprocessing

• Handling missing values
• Data normalization
• Cleaning transaction records

Feature Engineering

Key behavioral features are extracted from transaction data to improve model performance.

Model Training

Two machine learning models are trained for fraud classification:

• Random Forest
• XGBoost

Model Evaluation

Models are evaluated using:

• Accuracy
• Precision
• Recall
• F1 Score

The hybrid pipeline improved fraud detection performance by approximately 12% compared to baseline models.

---

Technology Stack

Machine Learning

• Python
• Scikit-learn
• Random Forest
• XGBoost
• Pandas
• NumPy

Backend

• Java
• Spring Boot
• REST APIs

ML Microservice

• Flask

Frontend

• React
• JavaScript

Tools

• Git
• Postman
• Linux

---

Dataset

The fraud detection models are trained using transactional datasets containing financial transaction records.

Dataset Characteristics

Typical dataset features include:

• Transaction amount
• Transaction timestamp
• Customer identifier
• Merchant information
• Transaction location
• Fraud label (fraudulent / legitimate)

Data Processing Steps

1. Removal of missing or corrupted records
2. Normalization of numerical features
3. Encoding categorical variables
4. Feature extraction for model training

These preprocessing steps help improve the accuracy and reliability of the fraud detection models.

---

Setup Instructions

1. Clone the Repository

git clone https://github.com/Zamishi/FraudDetection.git
cd FraudDetection

2. Setup Machine Learning Service

cd ml-service
pip install -r requirements.txt
python app.py

3. Run the Backend

mvn spring-boot:run

The backend handles API requests and communicates with the ML service.

4. Run the Frontend

cd frontend
npm install
npm start

---

Future Improvements

Potential improvements for the system include:

• Graph-based fraud detection techniques
• Explainable AI methods such as SHAP and LIME
• Real-time fraud detection for streaming transaction data
• Cloud deployment for scalable inference

---

Research Contribution

This project forms the implementation base for the research paper:

Hybrid AI Approach for Fraud Detection: Integrating Rule-Based Systems with Machine Learning and Graph Analysis

Accepted at ICAICCIT 2025.