Decision tree ensemble models are very practical for building predictive ML models. They are not strict on assumptions, can work on data without too much preprocessing, train fast and typically result in high accuracy. One of the weaknesses of these algorithms is that they are not designed for extrapolation.
There are quite many variations of decision tree ensembles, but which are more suitable to predict multiple time series? For a time t there are multiple observations, i.e. price of different products?
Here is a list of some of the algorithms:
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
The obvious answer is it depends on what you want to do.
AFAIK, all but one of the algorithms you've mentioned produce point forecasts. The one that's different is NGBoost as that produces probabilistic forecasts. If it's important to you to understand the uncertainty around your forecast, this could be a good option for you.
Even if you're only after a point forecast, NGBoost is a decent option in terms of performance but is significantly slower than the other ones you mentioned.
