Direct multi-step forecaster¶

This strategy, commonly known as direct multistep forecasting, is computationally more expensive than the recursive since it requires training several models. However, in some scenarios, it achieves better results. This type of model can be obtained with the ForecasterAutoregDirect class and can also include one or multiple exogenous variables.

Direct multi-step forecasting is a time series forecasting strategy in which a separate model is trained to predict each step in the forecast horizon. This is in contrast to recursive multi-step forecasting, where a single model is used to make predictions for all future time steps by recursively using its own output as input.

Direct multi-step forecasting can be more computationally expensive than recursive forecasting since it requires training multiple models. However, it can often achieve better accuracy in certain scenarios, particularly when there are complex patterns and dependencies in the data that are difficult to capture with a single model.

Direct multi-step forecasting can be performed using the ForecasterAutoregDirect class, which can also incorporate one or multiple exogenous variables to improve the accuracy of the forecasts.

Diagram of direct multi-step forecasting.

To train a ForecasterAutoregDirect a different training matrix is created for each model.

Transformation of a time series into matrices to train a direct multi-step forecasting model.