Python Feature Engineering Cookbook

Creating features from transactions with Featuretools

Featuretools is an open source Python library that allows us to automatically create features from time series and transactional databases with multiple transaction records for each specific entity, such as customers. With Featuretools, we can automatically create features at the transaction level. Such features include the day, month, and year from a datetime variable, the time between transactions, or if the transaction occurred on a weekend, as well as the cumulative sum or the difference in value between transactions.

Featuretools also aggregates existing and new features at the entity level—in our example, at the customer level—using mathematical and statistical operations, such as ...