You are predicting future employment rates from many economic features whose magnitudes differ greatly. To prevent features with large magnitudes from dominating the model, what preprocessing step should you apply before training?

Standardizing features (also known as Z-score normalization) transforms the data so that each feature has a mean of 0 and a standard deviation of 1. This is crucial when features have vastly different scales, as it prevents features with larger magnitudes from disproportionately influencing the model's learning process and distance calculations, which is common in many machine learning algorithms like gradient descent-based models or K-Nearest Neighbors. Quantile binning converts continuous data to categorical, which loses information and isn't primarily for magnitude scaling. A Cartesian product creates new interaction features, not for scaling individual feature magnitudes. Orthogonal sparse bigram (OSB) transformation is used for text processing, not for numerical feature scaling.