0
$\begingroup$

Let's imagine you have different models to give predictions on the same topic. One of your model is a regression, the other an ANN, the last one XGBOOST. Some of your models work better predicting at week+1, other at predicting week+3.

Each of your model has interesting results. How to best combine them into a strong one ?

My intuition is that you must feed the various forecasts to a new ensemble model, but also provide some of the initial features, so the ensemble model can learn in which case it is best to trust which original model.

I extensively red about Ensemble Modelling and it always seem that the only inputs are the forecasts of the base models, so no features.

Am I missing something or my intuition would be interesting?

$\endgroup$

1 Answer 1

0
$\begingroup$

What you are talking about is called Meta-Learning ("learning to learn"). And yes, those models learn from the output of other machine learning algorithms. If you have different predictions for different days into the future you might find a way to incorporate that into the meta learning model to weight the distinct model contributions over the given (prediction) time horizon.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.