Search code examples
recursionmachine-learningpredictionforecastingforecast

Forecasting a time horizon with machine learning in a recursive manner

I have a time series with hourly values that I predict for one day. My built up machine learning model has a single value as output. So that I have to run the machine learning model 24 times for a forecast of a whole day. For the prediction of time t+1, I give among other things the output of time t into the machine learning model (recursive structering).

Is there a special name for such recursive prediction models?

Solution

  • It sounds like you may be using a Direct-Recursive Hybrid Strategy. IE:

    hourly_prediction(t+1) = model1(obs(t-1), obs(t-2), ..., obs(t-24))
daily_prediction(t+2) = model2(hourly_prediction(t+1), obs(t-1), ..., obs(t-n))