Customer Attrition time line prediction using CoxTimeVaryingFitter without knowing the value of future co-variates

[naadvar]

Hi all,

I am trying to predict customer attrition by using a data set with one row per state change based on the docs, I have performance ratings that change over time and this leads to 4-5 records per employee.

I used the CoxTimeVaryingFitter to fit the model I had questions about how can I go about predicting the survival probabilities of the right-censored people beyond time =t without knowing all the values of the covariates in the future?

• Should I use the last know value to estimate the predict the partial hazard and calculate the survival probability that way?
• Or is the only option here to infer the values of the time-varying co-variates in the future and make a prediction.

If someone has dealt with this and could share an example that would be really helpful!