Appearance
Forecasts
Also known as
- Forecast a series
- forecast
- anomaly
- anomaly detection
- predict
- predictive analytics
- time series
- Holt-Winters
- trend
- seasonality
Forecasts and anomalies projects any time series forward and finds unusual points in it. The series can be a dataset counted or summed by day, week or month (orders, sign-ins, receipts …) or a KPI's history. Use it to anticipate volumes and to spot spikes that need a look.
Forecast a series
- Open AnalyticsForecasts (or press Ctrl K and choose Forecast a series). Stay on the Forecast tab.
- Choose the series:
- A dataset aggregated by date: choose the Dataset, the Date column, the Measure (Count, or Sum, Average, Minimum, Maximum of a Value column) and the Granularity (Daily, Weekly, Monthly).
- A KPI's series: choose the KPI.
- Optionally set From and To (Default: 90 days, 52 weeks or 24 months back).
- Set Horizon (periods): how many days, weeks or months to forecast (1–366, default 14).
- Leave Season length empty to derive it from the spacing (7 for daily, 12 for monthly), or type your own.
- Select Run forecast. The chart shows Actual, Forecast, Lower 95% and Upper 95%; the Model panel shows how well the model fits.
| Model field | Meaning |
|---|---|
| Method | Holt-Winters (additive) with enough history, Holt linear trend with less than two seasons, Naive (too few points) below three points. |
| Season length | The cycle used, for example 7 days. |
| Alpha (level), Beta (trend), Gamma (season) | The smoothing parameters, fitted automatically to minimise the one-step-ahead error. |
| RMSE, MAPE | The fit error in the series' unit and as a percentage. Lower is better. |
Find anomalies
- Select the Anomalies tab and choose the series as above.
- Set Window (points) (how many previous points each point is compared with, default 14) and Z-score threshold (how far from the rolling mean counts as unusual, default 3.0).
- Select Find anomalies. The chart shows the actual values, the Rolling mean and the band; points outside the band are listed with their Date, Direction, Value, Expected and Z-score.
The same analysis of the audit log appears on the Analytics overview, and forecast and anomaly widgets can be added to dashboards. Nothing on this page changes data.
If the series has no data, the page says The series has no data in this range.; widen From and To or choose another column.
Related
- Analytics overview
- KPI alerts
- Dashboards
- Dashboards and KPIs
- Plan to produce: demand forecasting in planning
