feature-engineering

## What I do - Create features from raw data for ML models - Transform variables for better model performance - Handle categorical and numerical features - Create interaction and polynomial features - Extract features from text, dates, and images - Handle missing data strategically - Select most predictive features ## When to use me Use me when: - Preparing data for ML models - Improving model performance - Working with raw/unstructured data - Handling complex data types - Reducing feature dimensionality ## Key Concepts ### Feature Engineering Techniques ```python import pandas as pd import numpy as np from sklearn.preprocessing import StandardScaler, OneHotEncoder from sklearn.impute import SimpleImputer # Numerical transformations df["log_income"] = np.log1p(df["income"]) df["income_squared"] = df["income"] ** 2 df["income_bucket"] = pd.cut(df["income"], bins=[0, 30, 60, 100], labels=["low", "mid", "high"]) # Date features df["order_date"] = pd.to_datetime(df["order_date"]) df["order_year"] = df["order_date"].dt.year df["order_month"] = df["order_date"].dt.month df["order_dayofweek"] = df["order_date"].dt.dayofweek df["is_weekend"] = df["order_dayofweek"].isin([5, 6]).astype(int) # Aggregation features user_stats = df.groupby("user_id").agg({ "order_id": "count", "total_amount": ["sum", "mean", "std"], "order_date": ["min", "max"] }).reset_index() # Merge back df = df.merge(user_stats, on="user_id", how="left") # Text features df["title_length"] = df["title"].str.len() df["word_count"] = df["title"].str.split().str.len() df["has_exclamation"] = df["title"].str.contains("!").astype(int) # Categorical encoding # One-hot encoding pd.get_dummies(df, columns=["category"], prefix="cat") # Target encoding mean_encoding = df.groupby("category")["target"].mean() df["category_encoded"] = df["category"].map(mean_encoding) ``` ### Feature Selection - **Filter methods**: Correlation, chi-square - **Wrapper methods**: RFE, forward selection - **Embedded methods**: LASSO, feature importance ### Handling Imbalanced Data - SMOTE oversampling - Undersampling majority class - Class weights in model - Focal loss