What is 'feature engineering' in machine learning?

Feature engineering often has more impact than algorithm choice. The techniques include Feature creation (debt-to-income ratio = total_debt/annual_income — more informative than raw values), Encoding (one-hot encoding for categorical variables like employment_type), Scaling (StandardScaler, MinMaxScaler — needed for distance-based algorithms), Binning (age groups), Log transformation (handling skewed income distributions), Date feature extraction (day_of_week, is_weekend for transaction fraud). In banking, creating features like "number of transactions in last 1 hour" or "velocity change vs 30-day average" are powerful fraud detection signals.