Survival analysis is a task dealing with time-to-event prediction. Aside from the well understood models, many others have recently emerged, however most of them lack interpretability as they are black-box models. Due to the unusual, functional type of prediction (either in the form of survival function or cumulative hazard function) standard model agnostic explanations cannot be applied directly.
The survex package provides model agnostic explanations
for survival models. If you’re unfamiliar with model agnostic
explanations, consider looking at the Explanatory Model Analysis e-book, as
most of the methods included in this package are extensions of those
described in the book for models with functional outputs.
The main function explain() creates a standardized
wrapper for a model, which is further used for calculating predictions.
If you work with models from mlr3proba,
censored, ranger, randomForestSRC
or survival packages, creating explainers is automated,
most often you only need to supply the model parameter to
the explain() function.
However, an explainer can be created for any
survival model, using the explain_survival() function by
passing model, data, y, and
predict_survival_function arguments.
The package can be installed from github using
devtools::install_github():
devtools::install_github("https://github.com/ModelOriented/survex")