Open in app
Alexandra Amidon

Sign in

Open in app
Alexandra Amidon

Jun 19, 2020·1 min read

The direct forecasting strategy uses a different ML model for each forecasting step. More specifically, each model is trained using as target the time series shifted of the desired number of time periods into the future. Imagine for example that one wants to train a 4 steps ahead model. In this case, each timestamp in t…

ML time series forecasting the right way
214
2

Mario Dagrada

Alternatively, if using a neural network model for forecasting, you can train a single network to output multiple predictions (direct forecasts) for each future time step. The network could then learn about relationships between the forecasts.

Alexandra Amidon

Data scientist working in the financial services industry