Step 3: Create a forecast model

To create a forecast model:

1. In Forecaster, select Forecast models > New forecast model.
2. Enter your Forecast model name.
3. Select a Data collection from the dropdown. 
4. Select an Algorithm from the dropdown.
   The list populates based on your chosen data collection. The algorithm you select trains the forecast model. See Algorithms to learn more.
5. Beneath Forecast horizon, use the dropdown to set the length of the forecast, select How many intervals.
   Forecaster determines the interval and maximum forecast length. It derives both the interval and length from the algorithm and the data collection. 
6. Beneath Additional settings, make the selections below (optional).
   • If the algorithm is Ensemble, select an Optimization metric from the dropdown. The default metric is MASE.  
   • Select from the Country-specific holiday calendar dropdown. This setting adds a related time series to the forecast model training. 
     • Country-specific holiday data is useful when the data collection represents data from a single country. 
     • To use a national holiday calendar, you must make sure the algorithm applied supports related data.
7. Map these fields: 
   • Workspace
   • Model
   • Module
   • Item ID list
   • Time dimension
   • Forecast (forecast version) list
   • Forecast list item
8. Select Create.
   You can check the status on the Forecast models page.

To generate the forward-looking forecast results, go to Step 4: Forecast Action.

Optionally, you can evaluate the forecast model before you create and run a forecast action. To evaluate:

• Assess metrics in the inspector panel tab.
• Examine backtest data and compare the forecast against the actuals.