In Google Cloud's Vertex AI Model Monitoring, the difference between these two concepts comes down to what you are comparing the current live data against.
This is a critical distinction for the Professional Machine Learning Engineer exam and real-world MLOps.
The Core Difference Training-Serving Skew: Compares Production Data vs. Training Data.
The Problem: "My model is seeing data in the real world that looks nothing like what it studied in the classroom."
Prediction Drift: Compares Production Data (Now) vs. Production Data (Past).
The Problem: " The world is changing. The data I see today looks different from the data I saw last month."
