Does anyone know of an R-package that will allow me to create an isotonic regression model , and not just output the new, corrected probabilities?
I have a random forest model that predicts the risk of mortality. I want to see if it can be improved further by adding a layer of model-calibration, and since the calibration plots are in no way sigmoid in shape, Im thinking isotonic regression is the way to go.
However, from all of the packages I have found so far that do isotonic regression (Iso, Stats and rfUtilities), I have not found one that actually outputs a model, that I can use on new data.
I want to train my rf model on the training set, calibrate it on a calibration set, and from those new calibrated probabilities/new model, test it again on a separate testing dataset.
Or am I missing something else entirely?? Any help is greatly appreciated!