Feature Engineering At Scale | ML System Design

Live Engine

Select Topic

easyFeature Engineering At Scale

A team builds a churn prediction model. One feature is "days since last login," computed as current_date - last_login_date. In training, current_date is the date the training pipeline runs. In production, current_date is the request timestamp. Three months after deployment, the model's recall on churned users drops significantly. What is the root cause?

Live Engine

Select Topic

easyFeature Engineering At Scale

A team builds a churn prediction model. One feature is "days since last login," computed as current_date - last_login_date. In training, current_date is the date the training pipeline runs. In production, current_date is the request timestamp. Three months after deployment, the model's recall on churned users drops significantly. What is the root cause?