search

MI2DataLab / survshap

SurvSHAP(t): Time-dependent explanations of machine learning survival models
https://doi.org/10.1016/j.knosys.2022.110234
MIT License
79 stars 15 forks source link

Expected Future Lifetime #30

Open hrafnfaedhir opened 8 months ago

hrafnfaedhir commented 8 months ago

I apologize if I am extrapolating beyond what your package was intended to do. I understand that a large portion of survSHAP(t) is providing explanatory variable analysis. Your paper indicates that SurvSHAP(t) can return an individual's unique survival function given their covariates, distinct from the assumptions in a Cox model. Can your Survival Function be used to calculate an Expected Future Lifetime? https://en.wikipedia.org/wiki/Survival_analysis#Quantities_derived_from_the_survival_distribution

krzyzinskim commented 8 months ago

Unfortunately, this is not the functionality of this package. SurvSHAP(t) returns the contributions of each covariate to the survival function prediction obtained by using another model. The fact is that summing these attributions gives the value of the survival function, but the prediction itself is from external model.

You can find some machine learning survival analysis here, for example: https://scikit-survival.readthedocs.io/en/stable/index.html.