Setting up continuous forecasting

This article describes how to set up continuous forecasting in the AI & Analytics Engine's Demand Forecasting Machine Learning Solution Template.

What is continuous forecasting?

Continuous forecasting in the AI & Analytics Engine is a feature that generates forecasts at regular intervals based on user-defined business requirements. This keeps predictions current as new data becomes available. The machine-learning models used for these forecasts are also re-trained when forecast quality drops below a desired threshold. By doing so, it ensures ongoing accuracy and up-to-date predictions without manual intervention, maintaining model performance as the data is updated.

How to set up continuous forecasting in the AI & Analytics Engine?

1. Select models to enable continuous forecasting:

To set up continuous forecasting, start by selecting the models to enable continuous forecasting.

 

Click Select Models

Choose the model(s) to enable continuous forecast

2. Configure input sources:

After selecting the model for continuous forecasting, the app summary page will show two options: Configure input sources and Provide data manually. These options offer flexibility in supplying data for ongoing forecasts:

• Configure input sources is suitable for users who have access to a database that is appropriately updated before each forecast's due date. This option automates the data refresh process, eliminating the need for manual intervention.

• Provide data manually is designed for scenarios where an automatically updated data source is not available, or users want to supply the missing data themselves.

In both options, the provided data must have the same schema as the datasets used when the app was created.

Update forecast by connecting to databases or providing data manually

 

💡The Provide data manually option is displayed only if the app was created some time ago and has an overdue forecast. A notification and guidance for providing the missing data to generate the new forecast will be shown.

After successfully connecting to the input source database, users can review the configurations by clicking Review forecast configuration in the App summary page.

3. Review forecast configuration:

On the forecast configuration page, users can customize the models used for continuous forecasting and adjust the criteria that triggers re-training when new data is updated. By default, models will be re-trained if their forecast accuracy falls below 80%. Furthermore, users can specify where the periodic forecasts should be sent, such as exporting them to a different database. They can also change the input data source if necessary.