💳 Fraud Transaction Detection

📌 Project Overview

This project implements a machine learning pipeline for detecting fraudulent financial transactions. The system analyzes transaction attributes such as transaction amount, customer behavior, and terminal activity to classify whether a transaction is legitimate or fraudulent.

Fraud detection is a critical application of machine learning in financial systems, helping institutions detect suspicious activities and prevent financial losses.

---

🎯 Objective

The goal of this project is to build a machine learning model capable of accurately identifying fraudulent transactions using transaction data.

The system performs:

• Data preprocessing
• Exploratory data analysis
• Feature engineering
• Model training and tuning
• Fraud prediction

---

🗂 Dataset

The dataset used in this project is a simulated financial transaction dataset containing both legitimate and fraudulent transactions.

Fraud scenarios included in the dataset:

1. Transactions above a certain threshold amount are marked as fraudulent.
2. Fraudulent activity originating from compromised payment terminals.
3. Customer accounts performing abnormal spending behavior due to leaked credentials.

Key dataset features:

Column	Description
TRANSACTION_ID	Unique transaction identifier
TX_DATETIME	Date and time of transaction
CUSTOMER_ID	Unique customer identifier
TERMINAL_ID	Merchant terminal identifier
TX_AMOUNT	Transaction amount
TX_FRAUD	Fraud label (0 = Legitimate, 1 = Fraud)

Due to size limitations, the dataset is not included in this repository.

---

🧠 Machine Learning Models Used

Several models were trained and evaluated:

• Logistic Regression
• Random Forest
• XGBoost

The best performing models were saved for deployment.

---

📦 Pretrained Models

Included in Repository

models/best_xgboost.pkl

Large Model Download (Random Forest)

Due to GitHub's file size limit (100MB), the Random Forest model is hosted externally.

Download here:

https://drive.google.com/file/d/1_HhKgvmPDDnjKwT2R4RiLN_ZeHjSxnhF/view?usp=sharing

After downloading:

1. Extract the ZIP file
2. Place the model file inside: models/

Example: models/ ├── best_xgboost.pkl └── best_randomforest.pkl

---

⚙️ Installation

Clone the repository git clone https://github.com/shrashtimittal/fraud-transaction-detection.git

cd fraud-transaction-detection

Install dependencies pip install -r requirements.txt

---

🚀 Running the Project

You can run training or evaluation scripts from the src directory.

Example: python src/train_models.py

For predictions: python src/predict.py

---

📂 Project Structure

fraud-transaction-detection │ ├── models │ └── best_xgboost.pkl │ ├── src │ ├── analyze_results.py │ ├── app.py │ ├── data_loader.py │ ├── eda_plots.py │ ├── eda_split.py │ ├── evaluate.py │ ├── predict.py │ ├── save_best_models.py │ ├── train_models.py │ └── tune_models.py │ ├── requirements.txt └── README.md

---

📊 Key Insights

• Fraudulent transactions often show abnormal transaction amounts.
• Terminal based fraud patterns can be detected through transaction clustering.
• Customer behavior analysis improves fraud detection accuracy.

---

🔮 Future Improvements

• Real-time fraud detection pipeline
• Integration with streaming transaction data
• Deployment as an API service
• Advanced anomaly detection models

---

👩‍💻 Author

Shrashti Mittal
Machine Learning & AI Enthusiast