Zansa › Answers › Forecasting and data science
What is model drift and how often should I retrain a forecast?
Model drift is the decline in a model's accuracy because the data or relationships it learned have changed. Retrain on a regular schedule, such as monthly or quarterly depending on how fast your business changes, and also whenever monitoring shows accuracy or bias worsening beyond a set threshold.
There are two kinds. Data drift means inputs look different, for example a new product range. Concept drift means the relationship changed, for example customers react differently to discounts after a competitor enters.
More frequent retraining is not always better, as it can chase noise and make forecasts unstable. Test whether retraining improves backtest results. Many time-series models are refit with every forecast cycle, which is cheap, while heavier machine-learning models may be retrained less often.
Want this applied to your business? Zansa builds dashboards and forecasts for small and mid-sized companies worldwide. Start with a free intro call.
Book a call