A normalized, relational dataset makes it easier to perform feature engineering. Unfortunately, raw data for machine learning is often stored as a single table, which makes the normalization process tedious and time-consuming.