Forecasting has several different options than standard predictions. These are outlined here.
Select the field that indicates the time stamp of each row. This is necessary to build a model that shows changes over time.
For datasets that contain more than one sequence, select an ID field to generate forecasts for each subsequence independently. A subsequence must include at least five dates in the forecast.
How far you want the model to predict the future. Forecasts get inherently less accurate as they get further from their training data. It's best to continue to provide fresh data into forecasting models to keep this gap small. The default value is 30%, i.e., if you have ten months of training data, it will project forward three months.
Bucket data into discrete periods (days, weeks) to smooth data that may not be in regular intervals and improve model accuracy.