Performance is robust across different simulated situations
We started with 𝞭max= 2000,pmean = 0.0008, 𝛔 = 0.0008. 𝞫 = 1, 𝞪 = 100, n = 5000, m = 100. This created a dataset with 5000 patients and 100 clinical features. Unless otherwise specified for testing model robustness, these are the base parameters we used. We built in three commonly used algorithms for testing: ExtraTreeRegressor, linear regression and Support Vector Machines (SVM)15.
With the above starting point, we examined the behaviors of the model. With the increase of mean termination rate of the population, performance stayed strong. (Fig 3b-c, Fig S1a, Fig 2 ). The median error rate at pmean = 0.0008 for cumulative errors are 9.11%, 8.97%, 9.10% for ExtraTreeRegressor, Linear Regression, and SVM, respectively, compared to 7.89%, 8.15%, 7.47%, which are their respective errors at pmean= 0.0012 . Overall, we saw little variance when the termination rate of the population changes.