Designing Data Governance from the Ground Up

Use Feature Stores to Prevent Data Drift

If you’re using data to train machine learning models, you can use a feature store to catalog all available features. It gives data scientists a single place to find the raw data they need to transform it into features their machine learning models can use.^[44] These features can be offline and calculated as part of a batch job (aka average monthly spend) or online and essential to calculate in real-time (such as fraud detection). Frameworks like SQL underpin offline features, whereas tracking online data demands platforms like Kafka that can track it in real time.

No matter which type of tracking you need, feature stores give data teams a single place to go where they can find the data and metadata they ...