2
$\begingroup$

I have a dataset that contains some categorical info about my users (like gender and country), whether they have churned or not and what's the time from their subscription to churn.

I want to use survival analysis and estimate the time between subscription to churn. My data has both censored data and event data. I want to plot Kaplan–Meier curve for two cohorts of male and female. But the data is highly skewed towards female (90% of the data is female users and 10% male users). For more context: I'm using Lifelines library and its KaplanMeierFitter.

Should I first balance the data for this two cohorts and then draw the Kaplan–Meier curve? or do I need not to care about the class imbalance in survival analysis.

$\endgroup$
2
$\begingroup$

No, it doesn't matter. An analogy is to compare another summary statistics: the average. If you want to compute the average of the two cohorts, would you balance the data first? No, of course not. However, the standard error will larger in the group with less individuals - same with the Kaplan Meier estimate.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.