Feature Engineering for Machine Learning

import pandas as pd
import numpy as np
from sklearn.preprocessing import (
    StandardScaler, OneHotEncoder, 
    KBinsDiscretizer, FunctionTransformer
)
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
from feature_engine import (
    datetime as dt_engine,
    imputation as imp,
    encoding as enc
)

# Sample data
data = {
    'transaction_date': pd.date_range('2023-01-01', periods=100, freq='D'),
    'amount': np.random.normal(100, 20, 100).round(2),
    'category': np.random.choice(['A', 'B', 'C', 'D'], 100),
    'customer_age': np.random.randint(18, 80, 100),
    'is_fraud': np.random.choice([0, 1], 100, p=[0.95, 0.05])
}
df = pd.DataFrame(data)

# Create time-based features
df['day_of_week'] = df['transaction_date'].dt.dayofweek
df['is_weekend'] = df['day_of_week'].isin([5, 6]).astype(int)
df['month'] = df['transaction_date'].dt.month

# Create interaction features
df['amount_per_age'] = df['amount'] / df['customer_age']

# Define preprocessing steps
numeric_features = ['amount', 'customer_age', 'amount_per_age']
categorical_features = ['category', 'day_of_week', 'month']

preprocessor = ColumnTransformer(
    transformers=[
        ('num', StandardScaler(), numeric_features),
        ('cat', OneHotEncoder(handle_unknown='ignore'), categorical_features)
    ])

# Create a feature engineering pipeline
feature_engineering_pipeline = Pipeline([
    ('preprocessor', preprocessor),
    # Add more feature engineering steps as needed
])

# Apply transformations
X = df.drop(['is_fraud', 'transaction_date'], axis=1)
y = df['is_fraud']

X_transformed = feature_engineering_pipeline.fit_transform(X)

# Get feature names after transformation
numeric_features_transformed = feature_engineering_pipeline.named_steps['preprocessor']\
    .named_transformers_['num'].get_feature_names_out(numeric_features)

categorical_features_transformed = feature_engineering_pipeline.named_steps['preprocessor']\
    .named_transformers_['cat'].get_feature_names_out(categorical_features)

all_features = np.concatenate([
    numeric_features_transformed,
    categorical_features_transformed
])

print(f"Total features after transformation: {len(all_features)}")

from category_encoders import TargetEncoder

# Initialize target encoder
target_enc = TargetEncoder(cols=['category'])

# Fit and transform
df['category_encoded'] = target_enc.fit_transform(
    df['category'], 
    df['is_fraud']
)

from sklearn.feature_selection import SelectKBest, f_classif

# Select top 10 features
selector = SelectKBest(score_func=f_classif, k=10)
X_selected = selector.fit_transform(X_transformed, y)

# Get selected feature indices
selected_indices = selector.get_support(indices=True)
selected_features = [all_features[i] for i in selected_indices]
print(f"Selected features: {selected_features}")

from tsfresh import extract_features
from tsfresh.utilities.dataframe_functions import roll_time_series

# Create time-series features
df_ts = roll_time_series(
    df, 
    column_id="customer_id",
    column_sort="transaction_date",
    max_timeshift=30,
    min_timeshift=5
)

# Extract time-series features
features_ts = extract_features(
    df_ts.drop("is_fraud", axis=1),
    column_id="id", 
    column_sort="transaction_date"
)