Detecting Model Drift

Learn the concept of model drift and how to detect it.

Understanding model drift

When we create a model and test its accuracy, we measure its performance at a certain moment. Over time, however, things can change, and the model's predictive power may decrease. Model drift is this drop in model performance we can experience due to changes in context. Suppose we want to forecast overall sales for a company, and we create a model based on past data. Many things can change over time—demand, customer behavior, competitors, etc. These changes would make our model obsolete with time.

Get hands-on with 1200+ tech skills courses.