I am looking for multi-step forecast deterministically with a random forest. I am aware the random forest isn't the best model for multi-step forecasting however I need to do it for comparison reasons. I have looked into multi-step forecasting methods the two methods I have found that may somewhat suitable. The first "directly" for which a separate model is developed to forecast each over each time. This is obviously very computationally expensive; for this reason, it will not be used. The other method is to use a Multi-Input Multi-Output (MIMO) RF, which instead of returning a single classification like traditional RFs it will output a vector of classifications. These multiple-classifications will give the forecasts till maturity. The issue with MIMO is that for each instances I will be forecasting will differ in length. For example one of instances to be forecasted my have to be forecasted for 3 months and other for 14 months (as it expires after this date).
Another way I have thought of doing this is by pulling data from an ICE plot (the ICE plot of time till maturity). So this would give me each individual instances dependence on time till maturity, this could be used to forecast each instance. However I have not found how to pull the data from a plot I have only seen how to plot this data on python (using PartialDependenceDisplay.from_estimator(clf, X, features, kind='individual') ). does anyone have any ideas of how I could do this or different ways of forecasting?