featuretools

Why we included this project

Featuretools automates the tedious part of building predictive models: turning raw relational and event data into features. It applies deep feature synthesis to entity sets, so instead of hand-coding each column, you describe how your tables relate and let the library generate dozens or hundreds of candidate features. Because it tracks time, it can build things like 'average amount spent in the last 30 days' while respecting cutoff times, which keeps future information from leaking into training. That makes it a good fit for supervised learning on transactional data, event streams, and sensor logs, where it slots into a pandas/DataFrame workflow ahead of scikit-learn or similar frameworks. It is a mature library with solid documentation, aimed at data engineers and ML practitioners who want feature construction to be repeatable across projects.