🎓 CodSoft Data Science Internship — Task 5

Credit Card
Fraud Detection

A machine learning system that identifies fraudulent credit card transactions from a highly imbalanced dataset of 284,807 transactions using advanced classification algorithms and comprehensive evaluation metrics.

🐍 Python 🤖 Scikit-Learn 📊 Pandas 📈 Matplotlib 🎯 Classification

284K+

Transactions

Features

Models

0.17%

Fraud Rate

📋 Overview

Project Overview

An end-to-end machine learning pipeline for detecting fraudulent credit card transactions, handling extreme class imbalance with a fraud ratio of just 0.172%.

🎯

Objective

Build a robust classification model to identify fraudulent transactions among 284,807 records with only 492 fraud cases.

⚖️

Key Challenge

Extreme class imbalance — fraud is only 0.172% of data. Standard accuracy is misleading; specialized metrics and techniques are critical.

🏗️

Approach

Balanced class weights, stratified splitting, feature engineering, GridSearchCV tuning, and evaluation with ROC-AUC, F1, and Precision-Recall.

📊 Dataset

Dataset Details

Credit card transactions made by European cardholders in September 2013, containing PCA-transformed anonymized features.

📑 Dataset Properties

Source	Kaggle — MLG ULB
Total Transactions	284,807
Fraud Cases	492 (0.172%)
Legitimate Cases	284,315 (99.828%)
Features	31 columns
Imbalance Ratio	~578:1

🔤 Feature Description

Feature	Description
Time	Seconds since first transaction
V1–V28	PCA-transformed (anonymized)
Amount	Transaction amount ($)
Class	0 = Legit, 1 = Fraud

🔄 Pipeline

ML Pipeline

A systematic machine learning workflow from data ingestion to model deployment.

📥 Load Data

→

🧹 Preprocess

→

⚙️ Features

→

🤖 Train

→

🎯 Tune

→

📊 Evaluate

→

💾 Save

Engineered Features

⏰

Hour & Is_Night

Cyclical hour extraction and night-time flag for temporal patterns.

💰

Log_Amount & Bins

Log-transformed amount to reduce skewness and binned categories.

📐

V_Mean & V_Std

Statistical aggregations across all 28 PCA components.

⚡

V_Extreme_Count

Count of V-features with extreme values as anomaly indicator.

🤖 Models

Models Used

Four classification algorithms trained with balanced class weights to handle the extreme imbalance in fraud detection.

📈

Logistic Regression

Linear model with balanced class weights. Fast training, highly interpretable, and provides probability estimates. Strong baseline for binary classification.

Linear Fast Interpretable

🌳

Decision Tree

Non-linear tree-based classifier that captures complex patterns. Handles class imbalance natively with balanced weights. Easy to visualize and understand.

Non-Linear Visual Balanced

🌲

Random Forest

Ensemble of decision trees with bootstrap aggregation. Reduces overfitting, provides feature importance scores, and handles imbalanced classes effectively.

Ensemble Robust Feature Rank

🚀

Gradient Boosting

Sequential boosting method that builds trees to correct previous errors. Excellent for complex patterns and typically achieves high ROC-AUC scores.

Boosting High Accuracy Sequential

📊 Results

Evaluation Results

Comprehensive model comparison across multiple metrics with emphasis on Recall, F1-Score, and ROC-AUC for fraud detection effectiveness.

Model	Accuracy	Precision	Recall	F1 Score	ROC-AUC
Logistic Regression	0.9745	0.0609	0.9184	0.1143	0.9769
Decision Tree	0.9991	0.7317	0.7551	0.7432	0.8774
Random Forest	0.9996	0.9487	0.7551	0.8409	0.9519
🏆 Gradient Boosting	0.9995	0.8750	0.7857	0.8279	0.9723

* Results are approximate. Run python main.py to generate actual metrics.

🎯

Precision

Of all predicted frauds, how many are actually fraud? Minimizes false alarms.

🔍

Recall

Of all actual frauds, how many were detected? Minimizes missed fraud cases.

📐

ROC-AUC

Area under ROC curve. Measures overall model discrimination ability across thresholds.

📁 Structure

Project Structure

Clean, modular architecture following software engineering best practices.

Credit_Card_Fraud_Detection/ │ ├── dataset/ │ └── download_dataset.md │ ├── notebook/ │ └── Credit_Card_Fraud_Detection.ipynb │ ├── src/ │ ├── __init__.py │ ├── preprocessing.py │ ├── feature_engineering.py │ ├── model_training.py │ ├── evaluation.py │ ├── prediction.py │ └── utils.py │ ├── models/ │ ├── fraud_detection_model.pkl │ └── scaler.pkl │ ├── outputs/ │ ├── class_distribution.png │ ├── transaction_amount_distribution.png │ ├── correlation_heatmap.png │ ├── fraud_vs_legitimate.png │ ├── missing_values_heatmap.png │ ├── feature_importance.png │ ├── confusion_matrix.png │ ├── roc_curve.png │ ├── precision_recall_curve.png │ ├── model_comparison.csv │ └── predictions.csv │ ├── screenshots/ │ └── README.md │ ├── README.md ├── requirements.txt ├── main.py ├── index.html └── .gitignore

🧩 Code

Code Architecture

Modular design with separation of concerns for maintainability and reusability.

🧹

preprocessing.py

Data loading, missing value analysis, duplicate detection, class distribution, train/test split, and StandardScaler application.

⚙️

feature_engineering.py

Time features (Hour, Is_Night), amount features (Log, Bins), and V-feature aggregations (Mean, Std, Extreme Count).

🤖

model_training.py

Multi-model training pipeline, model comparison, GridSearchCV hyperparameter tuning, and model persistence.

📊

evaluation.py

Classification reports, confusion matrix, ROC curve, precision-recall curve, and feature importance visualizations.

🔮

prediction.py

Prediction generation, predictions CSV export with actual/predicted/probability columns, and model loading utilities.

🛠️

utils.py

Logging configuration, directory creation, plot styling, DataFrame I/O, and project path management with pathlib.

📈 Visuals

Generated Visualizations

All visualizations are generated automatically by the pipeline and saved in the outputs/ directory.

▶️ Run

How to Run

Get the project running in just a few simple steps.

📦 Installation

# Clone the repository
git clone https://github.com/yourusername/Credit_Card_Fraud_Detection.git
cd Credit_Card_Fraud_Detection

# Create virtual environment
python -m venv venv
source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# Download dataset from Kaggle
kaggle datasets download -d mlg-ulb/creditcardfraud
unzip creditcardfraud.zip -d dataset/
                

🚀 Execution

# Option 1: Run complete pipeline
python main.py

# Option 2: Jupyter Notebook
cd notebook
jupyter notebook Credit_Card_Fraud_Detection.ipynb

# Option 3: View landing page
open index.html
                

✅ main.py will automatically:

• Load & preprocess data
• Engineer features
• Train & compare 4 models
• Tune best model (GridSearchCV)
• Generate all plots & CSVs
• Save model & scaler

Credit CardFraud Detection