# Is there any way to artificially create a probability calibration for data coming from another model?

I have predictions, which come from a survival model, this model gives me very low probabilities, and I am not sure if they fulfill the real probability of the phenomenon.

For example, I calculate $$P\left( T\leq t+d \middle| T>t \right)$$ and the probabilities are very low (with $$d=180$$).

To summarize, I need these probabilities to be on average another number (let's say $$0.2$$). Is it possible to create an artificial calibration with only this number (the desired average) as the input?

I have thought of creating a vector of size $$n$$ equal to the size of that distributes $$X_i \sim Ber(p=0.2)$$ and assign its ones to the top $$np$$ probabilities and its zeros to the latest $$n(1-p)$$. Which would result in a table with a column of probabilities obtained with the survival model and another column with an $$0$$ or $$1$$ depending on the said probability.

After getting this table, I would simply use CalibratedClassifierCV from scikit-learn. Is this the correct way?

• What survival model are you using? Commented May 11, 2022 at 14:56
• I am using Cox Proportional Hazards model. Calculating exp(-predict("expected")) with time = d+180 and the same with time = d Commented May 11, 2022 at 17:08
• It sounds like you are looking for something like a Bayesian survival analysis to specify priors Commented May 11, 2022 at 18:22
• Yes. Sound suitable, do you have any recommended bibliography? Commented May 12, 2022 at 1:02
• The survival times should be distributed as $$Exp(\beta \cdot X)$$ to get constant rates.
– 42-
Commented Jun 28, 2022 at 0:49