Machine Learning models are used to solve or improve a multitude of predictive analyzes in many different fields. We can find a wide range of problems and models used, but one of the few elements that almost all share is that time is involved. From the moment a sale occurs or a customer abandons, the age of an open opportunity in a CRM or when predicting the evolution of a price, time is present.

Used to our advantage, it allows us to generate features with great predictive power and evaluate our models realistically. Misused, however, it can cause data leaks that will obfuscate our ability to understand if the model is working and usable in production. You know, with great power comes great responsibility.

In this session we will review Machine Learning scenarios where time plays a fundamental role and we will describe, with examples taken from real and production projects, techniques and important points to take into account when working with time in ML.