Fraud Detection for E-commerce and Banking

This project leverages machine learning to detect fraudulent transactions in e-commerce and banking, aiding in proactive security and risk management. The goal is to provide a robust fraud detection pipeline with explainability, deployment, and dashboard visualization for actionable insights.

---

Project Overview

Key Features

• Data Analysis & Preprocessing: Handling missing values, data cleaning, and feature engineering for fraud detection.
• Model Building & Training: Comparison of multiple models, including deep learning architectures (CNN, RNN, LSTM).
• Explainability: Interpretation using SHAP and LIME for feature influence insights.
• Deployment: API service for real-time fraud predictions via Flask, Dockerized for scalability.
• Dashboard: Interactive visualization of fraud insights using Dash.

---

Project Directory Structure

The repository is organized as follows:

• .github/workflows/: Contains GitHub Actions for CI/CD and automated testing.
• .vscode/: Development configuration for Visual Studio Code.
• fraud-detection-api/: REST API implementation for serving fraud detection models.
• fraud-dashboard/: Dash application for real-time fraud data visualization.
• notebooks/: Jupyter notebooks for data exploration, feature engineering, and model prototyping.
• scripts/: Scripts for data preprocessing, visualization, and model building.
• tests/: Unit tests for model integrity and data processing functions.

---

Installation

Follow these steps to set up and run the project locally:

1. Clone the Repository

   git clone https://github.com/epythonlab/fraud-detection.git
   cd fraud-detection

2. Set Up a Virtual Environment

   For Linux/MacOS:

   python3 -m venv .venv
   source .venv/bin/activate

   For Windows:

   python -m venv .venv
   .venv\Scripts\activate

3. Install Required Packages

   pip install -r requirements.txt

---

Project Tasks and Workflow

Task 1 - Data Analysis and Preprocessing

• Handling Missing Values: Imputation or removal of missing data.
• Data Cleaning: Removing duplicates and correcting data types.
• Exploratory Data Analysis (EDA):
  • Univariate and bivariate analysis.
• Geolocation Analysis:
  • Convert IP addresses to integers.
  • Merge Fraud_Data.csv with IpAddress_to_Country.csv.
• Feature Engineering:
  • Transaction frequency and velocity.
  • Time-based features (hour of day, day of week).
• Normalization and Scaling
• Encoding Categorical Features

Task 2 - Model Building and Training

• Data Preparation: Feature and target separation, and train-test split.
• Model Selection:
  • Classical models: Logistic Regression, Decision Tree, Random Forest.
  • Advanced models: Gradient Boosting, MLP, CNN, RNN, LSTM.
• Model Training and Evaluation:
  • Train on both creditcard and Fraud_Data datasets.
• MLOps:
  • Use MLflow for versioning, experiment tracking, and model comparison.

Task 3 - Model Explainability

• SHAP (SHapley Additive exPlanations):
  • Explain feature importance using SHAP summary, force, and dependence plots.
• LIME (Local Interpretable Model-agnostic Explanations):
  • Generate feature importance plots for individual predictions.

Task 4 - Model Deployment and API Development

• Setting Up the Flask API:
  • Serve models via Flask in serve_model.py.
• Dockerization:
  • Create a Docker container for the API with a Dockerfile.
  • Run the container with:

    docker build -t fraud-detection-model .
    docker run -p 5000:5000 fraud-detection-model

• Logging:
  • Use Flask-Logging to monitor requests and track predictions.

Task 5 - Dashboard Development with Flask and Dash

• Interactive Dashboard:
  • Visualize fraud insights (transaction count, fraud cases, geographic data).
  • Use Dash to create charts (line, bar) and summary boxes for fraud trends.
  • Set up a Flask endpoint to serve fraud data for the Dash front end.

---

Contributing

We welcome contributions to enhance the project:

1. Fork the repository and create a new branch.
2. Make changes with clear, descriptive commit messages.
3. Submit a pull request with a detailed explanation.

---