New model idea

[waudinio27]

Hello Tyler! I see that you added nice examples that I want to try in the near future.
I was thinking about the following model idea. An ARIMA-boosted decision tree that does the smoothing for all components but leaves away the trend and seaosonality to keep it simple, and after this does the forecast with a regressor for every step of the multivariate forecast to avoid vanishing the predictive power.

What do you think about this idea? Could this be integrated or could you guide me with building this? Are you using reddit - what is your nickname there? So what I mean is, if you could be a bit of a tutor and teacher for me, as you have a lot of skills?

Greetings

Matthias

[tblume1992]

Do you mean like passing an ARIMA for the trend estimator, a decision tree as the exogenous estimator and then predicting but only with the tree exogenous estimator? Probably not understanding too well!

I have been thinking about setting up a discord for the project so that would be the ideal place to do some trainings or tutoring!

[waudinio27]

Discord is a good idea! My idea is to convert all input data to arima format and than do the boosting with with a proper estimator for every step of the forecast period.