Feature Engineering for Fraud Detection

You have a transaction dataset with columns: `user_id`, `merchant_id`, `amount`, `timestamp`, `location`, `device_id`. Design a feature engineering pipeline for a fraud detection model. What features would you create? How would you handle real-time feature computation?

solution.py

import pandas as pd
import numpy as np

def engineer_features(transactions: pd.DataFrame) -> pd.DataFrame:
    """
    Engineer features for fraud detection.

    Args:
        transactions: DataFrame with user_id, merchant_id, amount,
                      timestamp, location, device_id

    Returns:
        DataFrame with engineered features
    """
    features = transactions.copy()

    # Your feature engineering here
    # Examples of feature types to consider:
    # - User velocity features (transactions in last N hours)
    # - Amount anomaly features (z-score vs user history)
    # - Behavioral features (new device, new location)

    return features