Project-21-Python-For-Data-Analysis-Credit-Card-Fraud

End-to-end Credit Card Fraud Detection Analysis on 284,807 transactions | Analyzing fraud patterns, time-based activity, amount outliers & building a banking fraud dashboard using Python, Pandas & Seaborn

💳 Credit Card Fraud Detection Analysis

An end-to-end Fraud Detection Analysis project on 284,807 credit card transactions, completed as part of the Python for Data Analysis course.

📌 Objective

Detect fraud patterns, identify high-risk time windows, analyze suspicious transaction characteristics, detect amount outliers, and build a comprehensive banking fraud dashboard.

📂 Dataset

• Source: CreditCardData.csv — Real anonymized European cardholder data (Sept 2013)
• Records: 284,807 transactions
• Fraud Cases: 492 (0.172% of all transactions)
• Features: Time, V1–V28 (PCA-transformed), Amount, Class

⚠️ Dataset not included due to file size (>25MB).
Download from: https://www.kaggle.com/datasets/mlg-ulb/creditcardfraud

🛠️ Tools & Libraries

Library	Purpose
NumPy	IQR calculations, numerical operations
Pandas	Data cleaning, groupby, time analysis
Matplotlib	Histograms, time series, dashboard
Seaborn	Boxplots, heatmaps, scatter plots

❓ Analytical Questions (Q1 → Q15)

#	Question
Q1	Fraud percentage of all transactions
Q2	Total fraudulent transaction count
Q3	Fraud vs Legitimate distribution
Q4	Average amount: Fraud vs Legitimate
Q5	Top highest fraudulent transactions
Q6	Transaction amount distribution (histogram)
Q7	Correlation heatmap (key features)
Q8	Hour-by-hour fraud activity analysis
Q9	Do fraud transactions have higher amounts? (Boxplot)
Q10	Top suspicious transactions (high amount + early time)
Q11	Detailed amount distribution by buckets
Q12	Relationship between Amount and Time
Q13	Fraud trend over time (3-hour windows)
Q14	Outlier detection using IQR method
Q15	Banking Fraud Dashboard (8-panel)

💡 Key Insights

• Extreme class imbalance: fraud = only 0.17% of all transactions
• Fraud peaks in late night/early morning hours (0–6 AM)
• V14 and V17 are the strongest fraud predictors among PCA features
• Fraudulent transactions have lower median amounts but larger maximums
• Amount outliers show significantly higher fraud rates
• Fraud activity is not uniform — spikes at specific time windows

🚀 How to Run

git clone https://github.com/ammarelsayed-2a/Project-21-Python-For-Data-Analysis-Credit-Card-Fraud.git
cd Project-21-Python-For-Data-Analysis-Credit-Card-Fraud
jupyter notebook "Project 21 CreditCardFraud.ipynb"

👤 Author

Ammar Elsayed — Python for Data Analysis | 2026
LinkedIn