Credit-card-fraud-Detection

Credit Card Fraud Detection Pipeline

📌 Project Overview

The Credit Card Fraud Detection project is an end-to-end Machine Learning pipeline designed to identify fraudulent credit card transactions. Credit card fraud is a massive problem in the Fintech, Banking, and Payments industries (e.g., Paytm, PhonePe, Razorpay). This project tackles the inherent high-stakes class imbalance problem, where fraudulent transactions are extremely rare compared to normal transactions.

🚀 Why This Project Matters (Future Impact)

As digital transactions grow, so do sophisticated fraudulent activities. This project provides a foundational blueprint for:

1. Financial Security: Protecting customers from unauthorized charges and securing bank assets from massive losses.
2. Real-time Anomaly Detection: The models developed here (like Random Forest and Isolation Forest) can be adapted for real-time streaming pipelines to block transactions instantly.
3. Imbalanced Data Handling: Techniques demonstrated here (SMOTE, Class Weights, Undersampling) are highly transferable to other rare-event detection domains like Medical Diagnosis (rare diseases), Network Intrusion Detection, and Manufacturing Defect Detection.

🛠️ Tech Stack & Tools Used

• Python: Core programming language.
• Pandas & NumPy: For Data Manipulation, Descriptive Analysis, and Preprocessing.
• Scikit-Learn: Core library for Machine Learning models, Preprocessing (StandardScaler), and Evaluation Metrics.
• Imbalanced-Learn (imblearn): For specialized sampling techniques (SMOTE for oversampling, RandomUnderSampler).
• Matplotlib & Seaborn: For Exploratory Data Analysis (EDA) and visualizing model performance (ROC-AUC, Precision-Recall curves).
• Jupyter Notebook (nbformat, nbconvert): For interactive data analysis, EDA, and Predictive Analysis.

📂 Project Structure

• download_data.py: An automated script to fetch the raw dataset securely from GitHub.
• main.py: The core pipeline script that handles data loading, preprocessing, model training, and saves evaluation visualizations to the output/ folder.
• create_notebook.py: A script that generates the Jupyter Notebook programmatically.
• Analysis.ipynb: A comprehensive, fully executed notebook that contains Descriptive Analysis, Interactive EDA, Data Preprocessing, and Predictive Analysis step-by-step.
• requirements.txt: List of all Python dependencies.
• output/: Directory where all generated plots (EDA & Evaluation Curves) are saved.

⚙️ How to Run the Project

Prerequisites

Make sure you have Python 3.8+ installed.

1. Install Dependencies

Clone or download the repository, then install the required packages:

pip install -r requirements.txt

2. Download the Dataset

Run the data acquisition script to fetch the creditcard.csv dataset.

python download_data.py

The dataset will be downloaded into a dynamically created data/ folder.

3. Run the Main Pipeline

Execute the full, end-to-end training and evaluation script:

python main.py

You will see precision, recall, and F1-score outputs in the terminal. ROC-AUC curves, Precision-Recall curves, and Feature Importance plots will be saved in the output/ directory.

4. Interactive Analysis (Jupyter Notebook)

To view the detailed Descriptive Analysis, Exploratory Data Analysis (EDA), and Predictive Modeling interactively:

1. Open Analysis.ipynb in VS Code, JupyterLab, or your preferred IDE to view the executed results.
2. If you want to recreate the notebook from scratch, run python create_notebook.py and execute the cells.

🧠 Models Evaluated & Strategy

We trained and evaluated three distinct approaches to handling fraud:

1. Logistic Regression (with SMOTE and Undersampling):
   • Strategy: We synthetically generated fraud cases (SMOTE) and undersampled normal data to give the model a balanced view.
   • Result: Extremely high Recall (catches almost all fraud), but lower Precision (high false-positive rate).
2. Random Forest Classifier:
   • Strategy: Utilized class_weight='balanced' within the ensemble to penalize misclassifications of the minority class heavily.
   • Result: The best overall performer. It provides an excellent balance of high Precision (very few false alarms) and solid Recall, making it ideal for production where customer friction must be minimized.
3. Isolation Forest (Anomaly Detection):
   • Strategy: An unsupervised approach that isolates anomalies based on the assumption that frauds are “few and different.”
   • Result: A strong baseline that doesn’t rely entirely on perfectly labeled historical fraud data.

📊 Key Evaluation Metrics

In fraud detection, standard “Accuracy” is misleading because 99.8% of transactions are normal. Instead, we focus on:

• Recall: Out of all actual frauds, how many did we catch? (Crucial for stopping financial loss).
• Precision: Out of all flagged frauds, how many were actually fraud? (Crucial for avoiding customer frustration with blocked cards).
• ROC-AUC & Precision-Recall Curves: Used heavily in this project to visualize the trade-offs between precision and recall across different thresholds.

This site is open source. Improve this page.