Logistic Regression Fraud Detection

This dataset provides a small but representative sample of anonymized financial transactions intended for building and testing fraud detection models.

Each record represents a single transaction, including:

• Transaction type (e.g., CASH_OUT, TRANSFER)
• Transaction amount
• Sender and receiver account balances before and after the transaction
• Fraud indicator flags

It is suitable for:

• Binary classification
• Anomaly detection
• Machine learning tasks related to financial security

---

Dataset Structure

Column Name	Description
step	Time step of the transaction
type	Type of transaction (e.g., TRANSFER, CASH_OUT)
amount	Amount involved in the transaction
nameOrig	ID of sender account
oldbalanceOrg	Sender’s balance before the transaction
newbalanceOrig	Sender’s balance after the transaction
nameDest	ID of receiver account
oldbalanceDest	Receiver’s balance before the transaction
newbalanceDest	Receiver’s balance after the transaction
isFraud	Target variable: 1 if fraudulent, 0 otherwise
isPayment	Indicates if the transaction is a payment
isMovement	Indicates if it involved a balance change
accountDiff	Difference in account balances (derived feature)

---

Class Imbalance Notice

Important:
This dataset is highly imbalanced — the number of fraudulent transactions (isFraud = 1) is much lower compared to non-fraudulent ones.
This reflects real-world financial data and may affect model performance if not handled properly.

To improve results, consider:

• Resampling techniques like SMOTE or undersampling
• Using evaluation metrics like precision, recall, F1-score, or ROC-AUC instead of just accuracy

---

Inspiration

This dataset can help you explore:

• How fraud differs from legitimate behavior
• Techniques to detect rare but critical patterns
• How to evaluate models fairly when fraud is rare

---