💳 Credit Card Fraud Detection

📌 Problem Statement

Credit card fraud detection is a critical problem in financial systems due to highly imbalanced data, where fraudulent transactions represent a very small fraction of total transactions. Traditional accuracy metrics fail in such cases, making recall and ROC-AUC more important.

---

📊 Dataset

• Source: Kaggle Credit Card Fraud Dataset
• Transactions: ~284,000
• Fraud cases: ~0.17%

---

⚙️ Approach

🔹 Data Processing

• Checked missing values
• Converted target variable to integer
• Handled class imbalance awareness

🔹 Feature Engineering

• Hour-based transaction buckets (Morning, Afternoon, Evening, Night)
• Transaction amount bands (Micro, Low, Medium, High, Very High)
• Log transformation for skewed amount distribution

🔹 Model Used

• Logistic Regression
• StandardScaler for normalization

---

📈 Results

Metric	Value
ROC-AUC	~0.96
Recall (Fraud)	~57%
Accuracy	~99%

⚠️ Accuracy is misleading due to imbalance ✅ Recall is prioritized to catch fraud cases

---

💡 Key Insights

• Fraud transactions are extremely rare → imbalance problem
• High-value transactions show higher fraud probability
• Night-time transactions show slightly elevated risk
• Model performs well in separating fraud vs legitimate transactions

---

🏦 Business Impact

• Enables early fraud detection
• Reduces financial losses
• Helps prioritize high-risk transactions
• Can be integrated into real-time fraud detection systems

---

🛠 Tech Stack

• Python (Pandas, NumPy)
• Matplotlib, Seaborn
• Scikit-learn
• Google Colab

---

🚀 How to Run

1. Download dataset from Kaggle
2. Place it in your Google Drive
3. Update path in notebook
4. Run all cells

---

👩‍💻 Author

Sumukhi Pandey Aspiring Data Analyst | B.Tech CSE GitHub: https://github.com/Sumukhi90/credit-card-fraud-detection