Confidence bounds around predicted survival curves from a RandomSurvivalForest

[Husain-Kapadia]

Is your feature request related to a problem? Please describe.
My team and I have been using the RandomSurvivalForest model successfully. However, we wonder, if we could predict bounds around this survival curve that could indicate of how confident we are about the predicted survival curve? For example, if the bound is narrow, then very confident and if the bounds are wide, not so confident.

Describe the solution you'd like
A survival curve with bounds around it to indicate uncertainty in predictions.

Describe alternatives you've considered
So far, we have considered 2 possible solutions:

1. To calculate an error (which could be RMSE or MAE) for every example in our test set between the predicted survival curve (that we obtain by training the model on the train set) with the actual survival curve derived from the Kaplan Meier Estimator on the test set. Once we have this error, we could use that to derive bounds around the final predictions on unseen data (assuming that our error on the test set is also representative of the error we would have for making predictions on unseen data).

2. To aggregate the survival curves of different survival trees in the random forest and the mean of the curves would be the final predicted survival curve and the variance (at each time step) in the curves could be used to calculate the bounds (at each time step).

However, these are just ideas with certain assumptions and we are unsure if these solutions apply fundamentally too. We would like to know what you think about this problem and if you have a solution in mind that you could share with us?

References and existing implementations
We have not come across any existing implementation for this problem.