feature_engineering.py

import pandas as pd

def perform_feature_engineering(df):
    """
    Takes the raw simulation data and creates ML-ready features.
    """
    df_engineered = df.copy()
    
    # Ratio: Discretionary vs Essential Spending
    # Essential = Food + Bills
    df_engineered['essential_spending'] = df_engineered['spending_food'] + df_engineered['spending_bills']
    df_engineered['discretionary_spend_ratio'] = df_engineered['spending_entertainment'] / (df_engineered['essential_spending'] + 1)
    
    # Utility Payment Delay Indicator (Categorical to Numeric)
    df_engineered['utility_delay_indicator'] = df_engineered['utility_payment_timing'].apply(lambda x: 1 if x == "Late" else 0)
    
    # Ratio: EMI to Salary
    df_engineered['emi_to_salary_ratio'] = df_engineered['loan_emi_amount'] / (df_engineered['monthly_salary'] + 1)
    
    # Ratio: ATM Withdrawals to Salary
    df_engineered['atm_withdrawal_intensity'] = df_engineered['atm_withdrawals_amount'] / (df_engineered['monthly_salary'] + 1)

    # Encode categorical purely in case (not strictly needed since we manually turn strings to numeric above)
    df_engineered = pd.get_dummies(df_engineered, columns=['utility_payment_timing'], drop_first=True)
    
    # Select columns to drop (identifiers)
    cols_to_drop = ['customer_id', 'essential_spending']
    df_engineered = df_engineered.drop(columns=cols_to_drop, errors='ignore')
    
    return df_engineered

if __name__ == "__main__":
    from data_generation import generate_synthetic_data
    df = generate_synthetic_data(10)
    df_features = perform_feature_engineering(df)
    print(df_features.head())