Zansa › Answers › Forecasting and data science
How should I treat outliers in sales history before forecasting?
First find out why each outlier happened. If it was a one-off event that will not repeat, such as a data error, a bulk order or a stock-out, correct or flag it. If it reflects a recurring event like a promotion or holiday, keep it and add it as an input. Never delete points without a reason.
Automatic outlier removal can erase real information, such as a genuine surge you should be ready for next time.
A sensible process:
- Detect unusual points with simple rules or charts.
- Investigate with people who know the business.
- Classify as error, one-off, or recurring.
- Adjust or flag accordingly and document it.
Robust methods such as using the median are less sensitive to extreme values and are a useful safeguard.
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